The problem with the megahertz is that it's only useful to compare the same chip at different speeds. Small differences in cache size, bus speed or cache latency makes the speed comparison irrelevant. Cell and Xenon are very different chips, in an architectural way, the same can be said of RSX and Xenos, so speed comparisons are misleading.
Theoretical numbers only give you a slight idea of the possible performance, but in the end, it's a compromise of budget and time. These, and the capability of the developers, are the factors that provide you the amount of performance you can squeeze of the machines.







