Here are a few reviews comparing the PhII X4 965 and its slower siblings to the i7 920 as well as its faster siblings:
Anandtech AMD's Phenom II X4 965 Black Edition review
The Tech Report AMD's Phenom II X4 965 Black Edition processor review
Xbitlabs AMD Phenom II X4 965 Black Edition: The Peak of Deneb Evolution review
AMDZone Phenom II X4 965 Black Edition review
Lostcircuits AMD Phenom II X4 965 BE review
(I chose these reviews just because they contain a wide variety of applications used in their benchmarking, helps widen-out the performance perspective across the differing classes of desktop applications)
I understand you explicitly titled the thread topic to be a performance comparison, not a price/performance comparison...but I'll add
this link just to highlight what it is that some of the other posters are hitting on regarding i7 920's price/performance to the PhII X4 product line.
Not having assembled the numbers to do anything closely resembling statistical analysis, just going by my gut impression from looking at graph after graph of benchmark data, I would say my impression is that
at stock a PhII X4 965 appears to perform on par with an i7 920 in general.
It outperforms even the i7 965 in a few apps, but then again it underperforms the i7 920 to a seemingly similar margin in an equally few number of apps as well. What you want to do as a consumer is determine if your app falls into one of those categories where the architecture of choice really does make a substantial difference in the performance you will extract from your computer.
Not going to say much about comparisons between overclocked cpus as no two cpu's overclock the same, and any reviews based on comparing overclocked results are going to be inherently limited by the specific cpu sample they received for review so it says little about the statistics behind what the typical consumer can expect from their specific cpu in an overclocked environment.
That's not to say overclocking data and reviews have no value, quite the contrary, just saying for the purpose of your thread's topic (as I have interpreted it to be) I don't see the merit of injecting OC'ed results into the evaluation matrix at this point.