What??? I really should read MrTeal Chili thread... This pretty much reads as a problem I have...

I have one of mine that has been giving me issues lately, and it sounds very similar to your problem. A little debugging of the VRM to read the logged faults show that it's hitting a current limit fault and the DC/DC is tripping off. It's also a great performer, 16 engines on all 8 chips. The problem was that my cooling was good enough that the temperatures stayed under the limits, and with every engine enabled and ramping up to the full voltage (1.15V) I was pulling more than 160A out of the power supply. It was hashing over 40GH/s, but it would trip the protections on the DC/DC and shut down. Limiting the voltage to 1.1V has dropped the hashrate to 39GH/s, but it also runs stable now.

As of right now there isn't an option to limit voltage. Below is a version of 14e compiled to limit the voltage to 1.1V instead of 1.15V. Give that a try.
https://www.dropbox.com/s/i4rj2k1m9vrj00e/Chili14e1V1.hex

Over the break for the next couple weeks I plan to get some changes to the firmware done to allow limits to be imposed on different variables like frequency setpoint and voltage through command extensions.

BTW- Not super useful for most people since you can't use it while hashing since the mining software has the port open, but the ZlX command (lowercase "L") will return the temperature of all the chips. If you're compiling your own cgminer, you might want to add in a way to read that.