The clock speed does not impress me. I'm more interested in the processing speed, which is usually measured in FLOPS.
500GHz is utterly meaningless if it takes 500 clock cycles to do a single integer addition.
Deeper pipelines do not necessarily equate to higher possible clock speeds. The extra complexity often imposes other limits on the speed. Not only that, but deeper pipelines are affected more by cache misses. Larger caches to counter this problem are both expensive and power-consuming, which equates to more heat and a lower maximum clock speed. Finally, a deep pipeline requires accurate flow control prediction to be effective. A mispredicted jump means that the whole pipeline has to be flushed and it will take many more clock cycles to refill than a short pipeline.
AMD's current crop of CPUs have relatively short pipelines running at low clock speeds with small caches. They still outdo Intel's Pentium 4s. Since their jump prediction can be simpler it is more robust and cheaper to produce. Cache memory is hellishly expensive stuff and small caches further reduce the price.
Pentium-M is, as I'm sure everyone knows, based on the Pentium 3, but brought up to date. It is more similar to AMD's architectures.
Deep pipelines have faced off against short pipelines, and they lost.
[edit]
A decent framerate for FEAR is an average of 40fps, which an XFX GF7800GTX, Athlon X2 4200 and 2GB of dual-channel 500MHz RAM can maintain happily with everything turned on and turned up, except for FSAA (only at 4x) because I can't see much difference at higher sampling levels.
Now I come to think of it, the graphics card is overclocked, but it came that way. This particular XFX card runs about 12% faster than stock speeds.
