Well, my last thread looked at cache bandwidth and memory bandwidth, especially for multithreading, from a theoretical perspective. So this time, I'd thought I post some real world performance comparisons.
The first link deals specifically with Core 2 Duo's advantage over Core Duo. In general on the exact same labtop, you'll see an average 10% speed up with an equivalent clocked Merom over Yonah. That basically means Merom gives a one speed grade advantage over Yonah, ie a 1.83GHz T5600 is equivalent to a 2GHz T2500. The 4MB Meroms will likely see nearly another speed bump from the doubled cache so I wouldn't be surprised if a 2GHz T7200 is nearly as fast on average as a 2.33GHz T2700.
So basically, we're seeing a 10% performance increase on Merom over Yonah despite it being on the same 667MHz FSB which brings up the now age old question of how much bandwidth does Core 2 really need? JumpingJack's research has shown not much at all, and GamePC has run benchmarks between Merom (667MHz FSB) and Conroe (1067MHz FSB) and they show that in most cases they offer nearly identical performance despite the large memory bandwidth difference. Specifically, you should look at the 2.33GHz T7600 and the 2.4GHz E6600 since they have similar clock speeds and cache sizes. (The E6400 and E6300 only have half the cache of their Merom equivalents, which is why the Meroms are actually faster than them). In fact, the performance of the 2.33GHz Merom on a 667MHz is so good that like it's desktop E6600 cousin, it actually offers FX62 class performance on a notebook.
The place where the 667MHz FSB does show it's limits is in file compression which is very bandwidth determinant. That's not to say faster FSBs don't offer advantages, just that the extra bandwidth generally isn't actually needed and any benefits are in latency. This all bodes well for 1067MHz Kentsfields of course, although I'd still be more comfortable with a 1333MHz FSB.