[citation][nom]ARMforLowCost[/nom]Emulation should be used to ease the transition form x86 to ARM, Just until the software is ported to the native ARM code! When the FAB process for ARM chips reaches or approaches that of Intel's chips, ARM's lower transistor count will always consume less power. Currently the one thing that is helping Atom chipsis Intel's advanced power gating, but this only helps lower the power usage when the chip is idle! At full load, less transistors = less power used! With too much Intel inside, there will be too much money out of everyone's pockets![/citation]
Unless there are serious architectural deficiencies, more transistors = more work done at full load. The fabrication process doesn't have as much of an impact as one might think. Intel employs a lot of unique and interesting techniques to maximize instruction throughput, at the cost of power but not power efficiency or die area. Lower die area does not necessarily imply lower power consumption because logic networks that aren't being switched consume very little power.
Some architectures are inherently biased towards floating point instructions while others are inherently biased towards general purpose integer instructions. Intel owns the general purpose integer field, IBM owns floating point, while AMD owns the floating point / mm2.