I had problems with 2 separate rigs with 8 GPU's each. The first one, powered by a rosewill lightning 1300w, with 4 5970's had intermittent red light of death underneath the 5970, signalling bad power or pwm. I redid the cabling, giving each 8 pin pci power their own dedicated plug and the remaining 4 6 pin came from the remaining 2 available 6 pin power plugs; I used 2 of these
http://www.newegg.com/Product/Product.aspx?Item=N82E16812198018 to split those into 4 6 pins. Havent had any problems since. Motherboard Asrock 890gx extreme4 with Athlon II x2 250.
The other rig, also with 8 gpu (4 5870, 2 5970) could not reliably power more than 6 GPU's. It was powered by a PC Power & Cooling 1KW-SR (1000 watt continuous). Unfortunately by todays power standards, it gets pretty bad voltage droop under high loads, so sometimes I would get driver crashes or complete system freezes. I've since replaced it with an OCZ ZX 1250 watt. No problems since, and much better power efficiency, about 970 watts with all gpu's at 1v. Doing 2.99 Ghash. MOtherboard msi z77a-g45 with celeron g530.
Also a more simpler option is, as stated before, is "overclock less, cool more". I run 100% fan and 1v on all rigs, so I'm limited by ambient air and quality of gpu cores. If I get a driver crash, I decrease core speed by 5 MHz. I keep doing that until it stabilizes and I forget about it.
All rigs have 2gb budget ram, run off 250v ac power, windows 7 x64, using ultravnc for remote control, bfgminer 2.9.5, driver 11.12, sdk 2.1 installed on top of whatever sdk comes with 11.12. I use "gpu-platform" : 1 in my conf file to select my 2.1 sdk. Hope I was able to help you out.