Well, those are nvidia supplied benchmarks, so they have to be taken with a bit of salt. However, the fact that they benched at multiple resolutions and settings settings is a good sign.
It is very doubtful that they benched with Cat 10.3 though, so you can expect them to lose 5-10% against HD 5870 right there. Batman and FC2 are pretty heavily skewing the results in favour of Fermi -- and surprisingly so is battleforge, which has always been an ATI game! Fermi is also beating ATI in Hawx(!) but 10.3 gave a massive increase in that game so they're probably pretty even.
The most disappointing part of these results are the Crysis numbers, which I remember being disappointed with the 5870's numbers as well when it launched. Perhaps Crysis just doesen't scale too well with better hardware???
ATI should be able to match the PR numbers quite closely with a mix of 10.3+ drivers, higher binned parts, and 2gb frame buffer. Coming out six months later with 50% more transistors it's obvious that ATI's engineers did a better job this round and that Nvidia has some catching up to do architecturally. At the very least this isn't another Geforce FX or R600 (although like the R600 the Fermi architecture might be able to taken to new heights in future iterations, time will tell!)
The dark horse here are the Nvidia drivers. The hardware might be six months late, but that doesn't mean team is also behind. In any case, Nvidia has always been good with putting out performance drivers, even later in a products lifetime. For ATI, Recent drivers have had big perf increases, but R8xx is still clock/clock and shader/shader slower than R7xx despite huge enhancements to interchip bandwidth, so there might be a lot of untapped potential still in there.
For consumers, this is good. Close performance should mean that both companies will be working hard on drivers, fighting each other with price cuts, putting out faster SKUs, and generally be working hard to get/keep the crown.
Facebook
Twitter