Possibly a brief undervolt condition that's not enough to trip the power supply but enough to keep VRMs from firing up properly and the chips error out because they're taking slightly too long to initialize. The 380 probably isn't nearly as busy inside as the big guys (likely active PFC and a bunch of other mess) so it might have better output capacitors since there's likely more room for them. I'd try the capacitors-wedged-in-the-6pin trick and see if that helps smooth things out.
Otherwise, really no idea why it might work on a low-end PSU but not a big sexy one.
Sidehack,
Any chance that it comming up all xxxxxxxx's at boot is a thermal overload protection? (the BFL jalapenos, would vary the number of engines enabled every time after reboot, depending on how the chips did as it booted up)
I have noticed the only way for me to get it to come up all oooooooooooooooo's (with any consistency) is to put it in the freezer for 5 minutes then take it out and plug it in. I have tried this three time in a row, and it worked no fail.
Also if I do a warm reboot, all xxxxxxxxxxxx's again
unfortunately I still can't get it into high speed mode(38ghs) (to be totally honest, I did fluke into high speed twice before,but not sure what i did) anyways not a good batting average in over 100+ tries