Your post and the multicore post started to make me imagine what could really be done with a multicore.

This is entirely true.

This is true, as well. The problem is that people do not understand that serial and parallel can be made to operate dependent on each other. A multicore CPU in general will have the flexibility to "store" instructions between the CPUs. So, CPU_1 does an add. CPU_1 quickly sends the data to CPU_2 and CPU_2 does a subtraction while CPU_1 is adding yet another byte. In this manner, one builds up a FIFO of machine cycles and effectively increases the clock rate of a processor. Ideally this would have to be done on a super fast silicon die so that data transfer does not interfere with actual instructions. The more CPUs one has at their disposal the higher the clock rate. I would imagine 4 CPUs would be like 4x the original clock rate if the silicon speed was negligible. There would have to be data buffers in between the CPUs because the machine cycles for a divide are different than an add. So, there are some kinks, but I think they could be worked out. The idea stems from paralleling multiple LVDS lines. Highly creative, but could it be done? If it did work, then parallel a separate multicore with the first multicore to gain the advantage you stated above. I believe the chip manufacture, Intel, would have to make special modifications to allow programming in a serial manner because
This can be solved with a data FIFO, or fast memory as memory speeds are exceeding clock rates. Chances are, they have not thought of the idea. Who needs that kind of speed anyways, right?