Yes, I do ACTUALLY use Stata -- on a G4 Digital Audio, a Powerbook, and a Pentium IV at work. While I know it has not been optimized on a G5, Stata themselves state that it benefits from the dual FPU units on the G5. I encourage you to take a look at comments on the Stata listserv that refer directly to this, such as
this one. There is no benefit to optimizing Stata for Altivec, however. Single-precision Altivec wouldn't do it any good anyway. So on the G4 you're left with one FPU unit and no vector processing. Sort of Pentium III-ish, in other words.
Given that neither G4 nor the more limited G5 Altivec will do this software any good, I presume by lack of G5 optimization, you must be referring to lack of multiprocessing support? If so, I have a question for you -- is there a significant difference for non-SMP software between dual processors and a dual core processor?
The really depressing and sobering reality check part of your post is the news that Stata isn't even in the right compiling software for Universal Binary . . . . sounds like a project for Stata 10, I'm afraid. But the eternal optimist in me remembers how quick they were with Carbonizing Stata 7, a process completed and posted by August 2001 (and that was a mid-version upgrade, not a paid one).