Ars Technica has an interesting article on the "memory wall" that lies in the path of many-core glory:

Eight cores is the point where the memory wall causes a fall-off in performance on certain types of science and engineering workloads.

I'm sure we'll surmount this obstacle, but the solution might not come from Intel. A few months ago, I attended a talk given by a senior Intel engineer where he described the many-core future as inevitable. He said nothing, however, about future memory architectures and was unable even to speculate when asked. Intel's public road-map is silent on the problem, too.