SUCCESS with EVGA e-GeForce 8800GT 512-P3-N802-AR. It has a 1024k x 1S flash chip. I got this from Amazon.com. Green PCB.
It looks like the NVIDIA reference design. This is a factory overclocked 650MHz card, but of course you lose the overclock (even in Windows) when you flash it with the Apple firmware. So it's running at 600MHz. I might try overclocking in software using RivaTuner once I'm sure the card is stable at stock speeds.
And the power cable I mentioned earlier works.
http://shop.ati.com/product.asp?sku=3280778
Ran the Crysis demo for about 10 minutes at 2560x1600, everything set to high except for shaders and shadows set to medium. No AA. It looks great and plays beautifully. Unless this game slows down A LOT later in the game, I have no idea why anyone would fork out another $250 to run SLI. It has no lag at max resolution using one card. This is with the standard Apple bootcamp drivers. And the fan is barely audible. I do notice more heat coming out the back of the case.
Switched over to OS X and ran Quake 4 v1.4.2, everything maxed out... although for some reason 1920 x 1200 is the highest resolution available. Again, it's liquid smooth.
By the way, this card requires my HP monitor's DVI cable to be screwed down very tightly in order to get a picture. I thought the card was DOA at first, because I wasn't seeing anything even though it flashed ok and the fan spun up. The stock ATI card works even if the cable is loosely attached.
My system is a 2.8 dual quad, 10GB ram, 750GB WD RE2 hard disk.