Activitypub does have many very small nodes, very few very large nodes, and few medium sized ones. A lot of this is people wanting to be on the same server that their friends are on (even if there aren't that many advantages to it), and just that structurally, running a largeish server is more cost- and time-effective per user.
It'll be interesting to (if that ever happens) see whether bluesky, once it actually becomes a decentralized system instead of a closed one, shakes out differently due to protocol differences, or broadly follows the same patterns.
It is also at the same time contrary to everything he‘s said before about how he would run twitter - hell, he called this account out specifically as one he would not ban because he is oh so committed to free speech.
It will also not be enforced against anyone except if you‘re sharing the location of Elon, of a billionaire buddy of his, or kick up enough of a stink that it‘s a PR issue.
The person doing the cleaning on our office floor always wears very large and conspicuous probably noise canceling headphones, while working. Guessing she has had one too many person trying to make awkward conversation while she was just trying to get the coffee cups in the wash as quickly as possible.
It presumably helps somewhat if you want to find sponsors, or locations to hold parties, if you can point at an official list on which you are and say "look! this is a real thing! important people think this is worth doing!"
Other than that, well, there really isn't a downside, so yeah, why not.
One nice thing about a federated system is that admin work is also distributed. I run a small server that has existed for a long time (and has knowledge of / federates to some extent with most of the fediverse as such), and my co-moderator and me have a very manageable administration workload (mind: you only have to moderate content that actually arrives at your instance, not all content that is posted everywhere ever). The Mastodon software (and most other fediverse software) has multiple tools for this specifically.
On our instance, we do not allow "untagged" (not hidden behind a content warning) pornographic content. If an instance allows this, we "silence" them - meaning that none of their content shows up on any public timeline anymore. People can still follow users from there, but that's their choice, and none of my business - and nobody is getting a facefull of tiddy if they don't specifically want it.
If an instance is generally acting in good faith, but allows images that I definitely do not want cached locally, another tool you have is "media reject" an instance - in that case, no media from that instance will be downloaded. Can still follow, but have to click through for images, and no thumbnails. We try to avoid this, but, you know.
The nuclear option is to just defederate from an instance entirely. We also try to avoid this, but it's a very effectice "ugh god I'm not dealing with this" tool.
edit: That's on a macro level, of course - we can also do the same thing to single users, and that's obviously preferred if it's just one idiot. a user submitting a report can also choose to forward the report the the originating server, which can be helpful to get a problem at the root.
Mostly, that not every instance has the same policy.
You can always join a place where the credo is "we connect with everybody", while people who want a very filtered experience can join places that are a lot more nonchalant about blocking other instances. A lot of people probably do (and want to be in a place that does) something in the middle.
You could even have multiple accounts on different instances that you use in different contexts. Most mobile clients support this more or less well, though the default web UI does not.
I sometimes wonder if there is a large disconnect between two types of twitter users - those that use it as a sort of "news-gathering" medium, and those that use it as a conversation medium.
Twitter, as a company, seems to assume that the former is the default. I don't know if the majority of their users are actually like that. The vast majority of my tweets is replies to other people, and for most people I know it is the same, but of course, self-selection would do that. The analogy as twitter being unrolled IRC is, for me, definitely how I've been thinking of it for years.
If I understand correctly, that would be under the theoretical assumption that you can perform computations on numbers that require infinitely many bits (or at least, likely, exponentially many bits) to represent in constant time, which is an assumption that you are not just generally allowed to make.
If the end goal is to reduce third-party tracking by ad networks, then actually loading the ads, nevermind fake-clicking, certainly seems terribly misguided, and that's putting aside that anything short of a proper click is probably going to trigger ad networks bot detection measures.
It'll be interesting to (if that ever happens) see whether bluesky, once it actually becomes a decentralized system instead of a closed one, shakes out differently due to protocol differences, or broadly follows the same patterns.