Thinking about the history of clock rates on recent desktop chips, I'm sure you could look up data points to show that clock rate goes down each time the number of cores (or width of parallel instructions) increases, then may largely "recover" with the next "process shrink." As there aren't yet any 22nm CPUs with more than 8 cores/16 hyperthreads (that I know of), it's premature to project trace size vs. clock rate into the future on the basis of current data.
If Nvidia is successful, it certainly looks like very large numbers of cores will be associated with clock rate decreases, but obviously, you can't compare clock rates between widely different architectures.
Power efficiency certainly will continue to grow in importance, not only for small portable devices, but also for Intel MIC and GPGPU type accelerators in desktops, where there is a practical total power limit to be spread over an increasing number of cores. Data centers nowadays certainly consider their power budget when planning increased capability.
