2.66 divide by 2.5= 1.064 so expect 6% better, but barefeats reports much better results. One excuse is that Xeons are based on the core2 architecture which in itself provides a 5 to 10% gain clock for clock over the normal core (totaling 12 to 17% so far). On top of that the xeon is the top of the line and has (sometimes in the past) shown slight advantages over the remaining family processors clock for clock by a small margin. Speedmark 300/262=14.5%–> 14.5/2.66*2.5=13.6% better clock for clock.
Another reason would be that the apps are better optimized (mmx,sse1,2,3 and 64 bit might also have some bearing) i would go as far as saying as its not unintentional. There is complicated formula ive been working on to explain anomalies in benchmarks and clock for clock scaling (for lack of a better name im calling it interprocess latency).
Bottom line results are promising to say the least. I want an excuse to want one so that i dont just want one without a reason, but i just want one even without a reason.
From the macworld benchmarks you see the quad2.5g5 and quad2.66xeon seem pretty evenly matched, however the price difference is 700$! And you get much better graphics. And the 3ghz xeon should do even better.
As i predicted the g5 quad is still by no means obsolete in terms of absolute performance, but you cant argue with the price performance ratio of the new systems, in those terms they trounce the g5. The default Mac pro is (processor wise) the best bang for buck, 300$ for a 33% difference (2 vs 2.66) is good. But 1100$ for 3ghz? not the most economic in terms of bang for buck vs the 2.66 but hey absolute power comes at a price.
Coming soom my take on wwdc06 announcments.