Something that's not mentioned here but may be relevant is the fact that Nehalem's (i7 or Xeon) support TurboBoost which will bump the clock speed of one or two active cores by one multi (166 MHz) when two or more cores are idle. That makes a 2.93 into a 3.1GHz machine for most games, and combined with tri-channel DDR3 with on-die memory controller should perform very nicely compared to a 3.2 Harpertown which is shackled with high-latency FB-DIMM DDR2 running through a discrete North Bridge memory controller.
The other relevant bit, is the fact that Harpertown uses legacy FSB technology for communicating with the North Bridge (which handles the PCIe lanes for the GPU as well as memory control) which can be a limiting factor in moving data back and forth between the GPU and CPU. Nehalem on the other hand, finally retired the FSB in favor of QuickPath... a much more robust and higher performing technology and something that will be relevant with gaming, particularly when moving large textures back and forth.
I actually think that if there's one workload that can justify the 2009 architecture vs. the 2008, it's gaming!
The other relevant bit, is the fact that Harpertown uses legacy FSB technology for communicating with the North Bridge (which handles the PCIe lanes for the GPU as well as memory control) which can be a limiting factor in moving data back and forth between the GPU and CPU. Nehalem on the other hand, finally retired the FSB in favor of QuickPath... a much more robust and higher performing technology and something that will be relevant with gaming, particularly when moving large textures back and forth.
I actually think that if there's one workload that can justify the 2009 architecture vs. the 2008, it's gaming!