So far, every 1GHz+ T-Bird I've owned has given up an extra 100MHz after 2 weeks of intensive use. Believe me, I try to squeeze out every drop of performance on day 1 with these chips, I give up, I try 2 weeks later and voila!, I get an average of 100 extra MHz. I would say this effect is due to an x-factor I call "burn-in"; I can't scientifically explain it but I do scientifically observe the phenomenon in every one of these cpu's.
ex: 1GHz@1.4-->1.5, 1.2GHz@1.4-->1.5, 1.33GHz@1.5-->1.6
BTW, I use a few different stages of stability testing. I begin by looping the UT fly-by for over 24 hours. That'll usually weed out systems with overextended settings pretty quickly. I use UT because that's what I play and it seems more sensitive to system instability than most other games. Next, I loop 3DMark2001 and look for hops, skips, and locks. After that I'm usually pretty satisfied. If I suspect something's up with the system and UT and 3DM2K1 haven't picked it up, I'll run Prime95. A word about that: Prime95 is not optimized for Athlon cpu's and can make a system fail which would otherwise be stable; I just use it because it'll cause a flaky system to crash within 15-20 minutes.
Just my $0.02.