Hi Erik,
No, no idea or numbers about the fraction that would benefit.
The "optimisation" in TCO is different from its traditional sense, which is usually "just" to provide a constant factor improvement in speed/memory. TCO by contrast, allows or doesn't allow a programmer to use a certain programming technique: recursion.
I use recursion a lot, and when I moved to ABCL, I found that a lot of code broke on lists as short as a few thousand elements.
So I'm not referring so much to ABCL or its implementation as to a Lisp program. A portable Lisp programmer can't rely on TCO being available. So using defun/tro [1] lets them remain portable across Lisps.
It's probably not too hard to also optimise mutual recursion (f->g->f) in a labels clause using a similar macro to defun/tro.
When I say I'm not referring specifically to ABCL, it means that if ABCL implemented TCO, everything I wrote would just move over to another (present or future) non-TCO ANSI Common Lisp's discussion forum when someone enquired about it :-) .
Sorry about the delay. I haven't yet mastered this newsgroup's (mailing-list's?) interface, and thought my last message hadn't even gone through.
[1]: was deftro
Vibhu