I think that it can help to nudge individual weak cores towards stability when you get to the very edge of what the silicon will give you.
At least I thought it did help when I was messing with a Phenom II X4 which would fail one thread of Prime95 with the others happily chugging along seemingly indefinitely.
I'm fairly certain that I wasn't imagining the whole thing at the time
Overall, the rewards for tweaking ACC on Phenom II, mesured in performance_gained/time_spent seem pretty dismal -- apart for core-unlocking of course, which was an entirely other level of good-times.