The overblown frequency vs. cost efficiency trade-off

The amount of time you spend on things like clock skew and setting up pipeline flip flops is generally fixed per cycle. So, let's say that you need 50ps to cover clock skew and 100ps to cover flip-flop overhead. At 1GHz, you have 850ps to do work over 1ns (one clock cycle in this instance). At 2Ghz, you have 500ps-100ps-50ps or 350ps per cycle which means that you have 700ps to do work in 2 cycles at 2GHz.

At 1GHz, you have 850ps of work over the same time that you only have 700ps of work at 2GHz. That's quite significant.

There are other advantages at 1GHz as well. You burn far less power trying to keep your clock aligned since you can have a little more slack and still come out ahead. You also don't have to flop as many intermediate results. etc.

The problem is that we have only recently started caring about amount of work per (watt/area/cost/etc) instead of simply amount of work (total).