My previous graph is from the 8800 roundup on Bit-tech hardware reviews.
However I have also found the following: (a much newer review of a 8800Ultra that shows 8800GTX sli scores against a HD2900XT CF setup)
Company of Heros
The bottom graph (for some reason I can't get the graph to load as an image here) clearly shows the 8800GTX SLI setup scoring above 90fps with 4xAA and 16xAF at a resolution of 2560*1600. Considerably more than the 37fps per the op post benchmarks. (its also higher than the HD2900XT CF setup)
Again the bottom graph is the one to look at. This time its 65fps per the Op benchmarks and 88.5fps for the Bit-tech review (the latter also includes 4xAA and 16xAF whereas the Op benchmarks don't state whether any is running)
There are no Synthetic benchmarks run to compare 3dmark06.
This is running on Vista Ultimate with the following setup:
Intel Core 2 Extreme X6800 (operating at 2.93GHz -- 11x266MHz); Asus Striker Extreme motherboard (nForce 680i SLI); 2 x 1GB Corsair XMS2-8500C5 (operating in dual channel at DDR2-800 with 4-4-4-12-1T timings); Seagate Barracuda 7200.9 200GB SATA hard drive; Enermax Galaxy DXX 1000W PSU; Windows Vista Ultimate x86; Nvidia nForce standalone drivers version 15.00 WHQL.
So this system scores going on nearly 3x as high as the one in OP post despite being on a slower system overall (stock X6800 vs overclocked Quad core).
Thats says to me something was wrong with the SLI setup in the OP post benchmarks as the difference in results just isn't right
Also I take back my comment that the nvidia vista drivers suck if the bit-tech review is anything to go by they arn't too bad.
One thing I will note however from the bit-tech review is just how well the HD2900XT scales with CF in comparison to either the 8800GTS and the 800GTX at higher resolutions.