I thought that at first, and that could indeed be one thing he's saying. But I also agree with jmount that he probably understands that any tinkering is likely to produce as many bugs as it fixes, and therefore implementation changes are to be avoided as well unless addressing actual problems that people are experiencing.
Yeah it is interesting. My read is he is saying that any de-stabalizing change isn't worth it at this point (even if it is a net improvement). Yet we all envision Knuth as an optimizer and tinkerer. Of course few people have experience maintaining a continuously important and popular tool for such a long interval.
