Swinging type-bars in general weren't a problem; it was the kind of swinging type bars used in (later) "manual" mechanical typewriters that were. (And also, that's not what I mean by "buffered.")
The oldest (late 1800s) mechanical typewriters were designed simply, with type mechanisms engineered under the assumption of a slow hunt-and-peck typist who would only be pressing one key at a time.
And this worked well, until typing as a skill developed, and people began to activate keys concurrently (with one type-bar potentially beginning to swing forward, before a neighbouring type-bar finished its return.) This caused the neighbouring type blocks on the end of the bars to end up binding together — usually at the end of their bars' travel, causing both keys to remain stuck down.
According to folk history, this type-bar-binding problem was solved in part through the development of the QWERTY layout — moving frequently-successive English-language letter-pairs so that they wouldn't be done using neighbouring keys. Regardless of the veracity of that, this still didn't truly solve the problem — early QWERTY typewriters still end up binding for many common words.
Later mechanical-typewriter manufacturers sought to solve this problem "for real", through internal engineering tweaks. Look up patents from the early 1900s-1920s, and you'll see many of these proposed. But one two-part approach stood out, and became nearly-universally adopted:
1. chamfering and carbon-lubricating the edges of the type blocks on the end of the type-bars; and then
2. allowing the type-bar linkages a bit of horizontal travel.
These two tweaks together, allowed the type blocks on the end of the type-bars to slip past one-another (at least when travelling at differing velocities, or with one rising and the other falling.) And this virtually eliminated binding — but not without trade-offs.
Mechanical typewriters with these fixes, have a slight looseness to the type-bars, and so a slight random wobble in the horizontal placement of type on the page (especially when the key was depressed slowly). And if two neighbouring keys are depressed in quick succession, either one or both keys are liable to shove the other slightly, resulting in one having quite a bit of horizontal displacement, and even a slight "edge-on impact" where one side of the type doesn't get fully inked onto the page. These characteristics can be easily observed in type specimens from the early 1900s, to the point that they are often used to date such specimens.
Nobody wanted to use these typewriters for professional typesetting if they could help it; they were just too "sloppy" at that point. (The earlier designs weren't, but the typebar-clash problem was just too bad for any "modern" professional typist of the era to volunteer to use them.)
The introduction of electromechanical typewriters in general, allowed large changes to the type-bars and type, for the same reason that drive-by-wire in cars allowed large changes to the transmission and linkages: without a sloppy human in the circuit, things could be far more crisp. Key-clashes could be solved without mechanical slop; and so the mechanism could be tightened back up.
At first, electromechanical typewriters — having nothing like a memory — were instead wired in a blocking fashion to avoid key-clashes. If you pressed two neighbouring keys in quick succession, the second key would "jam", refusing to depress until the first one returned. Essentially, all neighbouring pairs of keys were interlocked. This made typing on them very annoying — but required no fiddling to unstick, just a very attentive approach to typing, involving watching/feeling for unstruck keys, immediately stopping to re-key them. And the results were very crisp. So these typewriters were one of the first "enterprise hardware" — forced on typists (who didn't like them, but did tolerate them) by companies who wanted clean typed reports.
The next natural step was to turn the blocking solution into a non-blocking / asynchronous one. This was done through one-key buffering — with the typewriter using two loops of analogue(!) https://en.wikipedia.org/wiki/Delay-line_memory as registers, to hold alternating key actuations and then play them back out. For most key actuations, this replay was instantaneous, with the electrical linkages to the type-bars being actuated concurrently. But neighbouring keys, again through electrical interlocks, would "put a hold on" their sibling key's replay (if that was the key struck) until their own replay was done. (One could compare/contrast these typewriter mechanisms with pinball-machine mechanisms, both being invented around the same time.)
When digital electronics came along, the delay-line memories were now storing digital serial codes instead of parallel electrical voltages; and the number of bits able to be buffered and refreshed in the delay-line memory increased, so multiple keypresses of neighbouring keys, or the same key, could be buffered as fast as they could be pressed, and then replayed in sequence. (Some early computer TTYs were built using this sort of typewriter platform!)
Some electronic typewriters in the late 1960s also switched from delay-line memory to the just-invented shift-register memory — which made them much more accessible to the mass-market, rather than only institutional buyers.
And now we get to the original IBM Selectric Composer: it used magnetic tape for memory — and didn't immediately replay it, instead buffering indefinitely until you manually trigger replay.
Which, mind you, isn't what made the Composer really good for printing. The thing that did that, was that the buffering allowed it to measure text and then typeset it using proportional-width type. But this actually required you to type everything twice (in immediate mode) or for it to read the magnetic tape twice (in buffered mode) — so really, the advancement here wasn't the memory, it was the compute. The ability of the Composer to accumulate the widths of keypresses, given a loaded table of widths of glyphs for the font in use, into an accumulator register, was what made "online" proportional typesetting possible.
The Composer's use of external magnetic tape, also enabled the use of a cheaper sibling (the https://en.wikipedia.org/wiki/IBM_MT/ST) to input the text and test-print it monospaced; while the Composer would just be used for proportional-type replay. In this use-case, the Composer itself was essentially playing the role of a typesetting machine, rather than a typewriter.
But in terms of printing monospace text cleanly, the Composer wasn't any big advancement. Everything above since the first blocking-key electric typewriter was printing monospace text cleanly. Look at type from (expensive, niche) electrical typewriters from the 1930s on — they'll be crisp†. Telegrams from the period were printed on such machines, and look great.
† Well, okay, there'll be some bleed; those ultra-crisp type edges did indeed only come along with polymer ink ribbons, and those were introduced/popularized along with the first Selectrics. But — presuming a fresh cloth ink ribbon — the type will always be fully inked, square on the page, not smudged, no wobble, no unwanted horizontal offset, etc. Mechanical-linkage typewriters had all of those problems!
It's just that nobody really cared about any of that (other than those companies wanting really clean-looking reports), since nobody was doing offset-lithography mastering using monospace typefaces.
It's only once the marketable use-case changed (with the Composer) that people really started to notice that typewriters had slowly become as good at producing a nice typeset page as a professional publishing house was.
The oldest (late 1800s) mechanical typewriters were designed simply, with type mechanisms engineered under the assumption of a slow hunt-and-peck typist who would only be pressing one key at a time.
And this worked well, until typing as a skill developed, and people began to activate keys concurrently (with one type-bar potentially beginning to swing forward, before a neighbouring type-bar finished its return.) This caused the neighbouring type blocks on the end of the bars to end up binding together — usually at the end of their bars' travel, causing both keys to remain stuck down.
According to folk history, this type-bar-binding problem was solved in part through the development of the QWERTY layout — moving frequently-successive English-language letter-pairs so that they wouldn't be done using neighbouring keys. Regardless of the veracity of that, this still didn't truly solve the problem — early QWERTY typewriters still end up binding for many common words.
Later mechanical-typewriter manufacturers sought to solve this problem "for real", through internal engineering tweaks. Look up patents from the early 1900s-1920s, and you'll see many of these proposed. But one two-part approach stood out, and became nearly-universally adopted:
1. chamfering and carbon-lubricating the edges of the type blocks on the end of the type-bars; and then
2. allowing the type-bar linkages a bit of horizontal travel.
These two tweaks together, allowed the type blocks on the end of the type-bars to slip past one-another (at least when travelling at differing velocities, or with one rising and the other falling.) And this virtually eliminated binding — but not without trade-offs.
Mechanical typewriters with these fixes, have a slight looseness to the type-bars, and so a slight random wobble in the horizontal placement of type on the page (especially when the key was depressed slowly). And if two neighbouring keys are depressed in quick succession, either one or both keys are liable to shove the other slightly, resulting in one having quite a bit of horizontal displacement, and even a slight "edge-on impact" where one side of the type doesn't get fully inked onto the page. These characteristics can be easily observed in type specimens from the early 1900s, to the point that they are often used to date such specimens.
Nobody wanted to use these typewriters for professional typesetting if they could help it; they were just too "sloppy" at that point. (The earlier designs weren't, but the typebar-clash problem was just too bad for any "modern" professional typist of the era to volunteer to use them.)
The introduction of electromechanical typewriters in general, allowed large changes to the type-bars and type, for the same reason that drive-by-wire in cars allowed large changes to the transmission and linkages: without a sloppy human in the circuit, things could be far more crisp. Key-clashes could be solved without mechanical slop; and so the mechanism could be tightened back up.
At first, electromechanical typewriters — having nothing like a memory — were instead wired in a blocking fashion to avoid key-clashes. If you pressed two neighbouring keys in quick succession, the second key would "jam", refusing to depress until the first one returned. Essentially, all neighbouring pairs of keys were interlocked. This made typing on them very annoying — but required no fiddling to unstick, just a very attentive approach to typing, involving watching/feeling for unstruck keys, immediately stopping to re-key them. And the results were very crisp. So these typewriters were one of the first "enterprise hardware" — forced on typists (who didn't like them, but did tolerate them) by companies who wanted clean typed reports.
The next natural step was to turn the blocking solution into a non-blocking / asynchronous one. This was done through one-key buffering — with the typewriter using two loops of analogue(!) https://en.wikipedia.org/wiki/Delay-line_memory as registers, to hold alternating key actuations and then play them back out. For most key actuations, this replay was instantaneous, with the electrical linkages to the type-bars being actuated concurrently. But neighbouring keys, again through electrical interlocks, would "put a hold on" their sibling key's replay (if that was the key struck) until their own replay was done. (One could compare/contrast these typewriter mechanisms with pinball-machine mechanisms, both being invented around the same time.)
When digital electronics came along, the delay-line memories were now storing digital serial codes instead of parallel electrical voltages; and the number of bits able to be buffered and refreshed in the delay-line memory increased, so multiple keypresses of neighbouring keys, or the same key, could be buffered as fast as they could be pressed, and then replayed in sequence. (Some early computer TTYs were built using this sort of typewriter platform!)
Some electronic typewriters in the late 1960s also switched from delay-line memory to the just-invented shift-register memory — which made them much more accessible to the mass-market, rather than only institutional buyers.
And now we get to the original IBM Selectric Composer: it used magnetic tape for memory — and didn't immediately replay it, instead buffering indefinitely until you manually trigger replay.
Which, mind you, isn't what made the Composer really good for printing. The thing that did that, was that the buffering allowed it to measure text and then typeset it using proportional-width type. But this actually required you to type everything twice (in immediate mode) or for it to read the magnetic tape twice (in buffered mode) — so really, the advancement here wasn't the memory, it was the compute. The ability of the Composer to accumulate the widths of keypresses, given a loaded table of widths of glyphs for the font in use, into an accumulator register, was what made "online" proportional typesetting possible.
The Composer's use of external magnetic tape, also enabled the use of a cheaper sibling (the https://en.wikipedia.org/wiki/IBM_MT/ST) to input the text and test-print it monospaced; while the Composer would just be used for proportional-type replay. In this use-case, the Composer itself was essentially playing the role of a typesetting machine, rather than a typewriter.
But in terms of printing monospace text cleanly, the Composer wasn't any big advancement. Everything above since the first blocking-key electric typewriter was printing monospace text cleanly. Look at type from (expensive, niche) electrical typewriters from the 1930s on — they'll be crisp†. Telegrams from the period were printed on such machines, and look great.
† Well, okay, there'll be some bleed; those ultra-crisp type edges did indeed only come along with polymer ink ribbons, and those were introduced/popularized along with the first Selectrics. But — presuming a fresh cloth ink ribbon — the type will always be fully inked, square on the page, not smudged, no wobble, no unwanted horizontal offset, etc. Mechanical-linkage typewriters had all of those problems!
It's just that nobody really cared about any of that (other than those companies wanting really clean-looking reports), since nobody was doing offset-lithography mastering using monospace typefaces.
It's only once the marketable use-case changed (with the Composer) that people really started to notice that typewriters had slowly become as good at producing a nice typeset page as a professional publishing house was.