I guess what I'm trying to voice is my confusion over what I hear, which is a real desire for "lisp all the way down" in the community versus what I see, which is surface level lisp.
Well, if I had to guess the biggest part of it is practicality. This of course effects every large project, but there are a few things that make Lisp operating systems less practical than other OS projects. Three that come to my mind are that the community of Lisp programmers is far smaller than that of C programmers, the second being that the Lisp philosophy is opposed to half-baked work, and the third being that the Lisp ideas of the way a computing system should work are so different from the status quo that it is difficult to create something which is capable of accepting foreign applications, and building on existing infrastructure. This is just a guess though.
Am I to understand UNIX hackers try to write constant-time operations with fallbacks, which of course has the mentioned disadvantage that these can cascade and cause unacceptable performance?
I misspoke, this wasn't what I was trying to say. I was trying to point out that in the most pure form of the UNIX philosophy any sort of optimisation is just seen as complexity and so it's avoided. Rather the domain space is constrained to some subset of the full problem where a naive implementation is suitable. The UNIX hacker works in C and writes naive algorithms so they naturally write algorithms with good constant, but poor big-O performance. Rob Pike for example is known for his rule: “Fancy algorithms are slow when n is small, and n is usually small.”