Welcome to the new world of CPU physics -- clock rate doesn't tell you much other than how much heat is going to be created as a by-product of computing.
In physics, one of the few things I remember is "force applied over a distance is work". I remember it because it was always punctuated with "my students don't know what work is". Clock rates are force -- raw, unadulterated oomph. In the theoretical world, force is force, and the little block slides across the imaginary plane until it collides with the billiard ball -- never mind. The real world has such niceties as friction, which impair that force and cause the useful work to be reduced, resulting in heat and wear. In CPU terms, that friction is memory latency, and it's getting worse in relative terms. A generation of CPUs ago, you might have drawn the comparison to the slight but perceptible friction of an object on a sheet of ice. Today, that friction is modeled more closely by the same object on a sheet of asphalt. Force is great, but unless you remove the friction from the system you generate heat without doing a lot of useful work.
Commercial message: this is one reason why I'm still at Sun after 15 years. Our SPARC guys figured this out (using much better physics than me) and began solving for the rate limiting constraints: heat, space, and memory latency. Multi-core, multi-threaded chips are going to reshape the system design physics, and ideally change the single figures of merit we use to decide what systems really do the work of IT.