Consider this, the X850XT PE crushes the 6800GT. In tandem it will do the same
Assuming AMR is as good as NVIDIA's SLI. What I mean is, no one knows how efficient ATI's method will be compared to NVIDIA's method.
Down side is that AMR will be less efficient and R5x0 + R4x0 won't be as good as two 6800Us in SLI.
Middle ground is that they will be equal (win some lose some, but even number on each side).
Up side is that AMR might be more efficient than NVIDIA's SLI and R5x0 + R4x0 will be better than two 6800Us in SLI.
One thing to consider here is that NVIDIA has had a lot longer to work on their SLI now. G70 might very well have a better hardware implementation (more efficient in all or some respects).
If that is true, two 6800Us in SLI might be beat by an R5x0 + R4x0, but R5x0 + R5x0 might be beat by G70 + G70 in SLI (even assuming G70 and R5x0 are the same speed as each other in single card use).
I guess my real point is that you can't assume anything right now.
NVIDIA has a big lead on ATI in terms of design and tweak time (NVIDIA has the lead with their Dual Graphics Card Solution and with SM3.0 (leading with SM3.0 is significant because NVIDIA has had all this time to come up with a better solution. Same could be said for ATI since they have waited, but generally Version 2 of anything is better than Version 1 and ATI has no Version 1 yet)).