Comment 9 for bug 756926

string replace transforms for simple-base <-> simple-character were present as of https://sourceforge.net/p/sbcl/sbcl/ci/260a9146f023
so I'm not really sure what was being asked here- Maybe runtime determination of the best replace routine for compile-time-unknown element-type? I suspect we do that now.
And the initial comment is unclear - FOO2 is faster than FOO2 ???

At any rate, I found that for a string of about 100 mega-characters, all of FOO0, FOO1, and FOO2 perform roughly the same, certainly not 20x apart. And at that string length, memory cache behavior is the dominating factor. So I can't really say what we do for short strings because they're too small to measure and I didn't bother to check.