You mean like Java and Android?
transmeta basically did this too, and well they are gone now.
i am not some cpu expert, but from what i've read the x86 decode logic on an intel chip hasnt exactly grown much since the first chips that translated to internal risc like operations. (which i think was the p6).
so basically the number of transistors dedicated to translating from x86 has remained the same while the rest of the chip has gotten bigger (so the piece of the die dedicated to it is smaller and smaller with each node size decrease).
that said, intel doesn't want to change instruction sets. intel has the best fabs, and the fastest cpus bar none. if they say switched the decoder out to say decode arm into say intel micro ops, they would be faster obviously (lets say this was possible).
but the whol epoint of staying on x86 is that they dont have as much competition and they have the advantage of backwards compatibility. without that they are just another builder of cpus, albiet a very good one. same with amd.
neither amd or intel want to compete with broadcom, qualcom, samsung, TI etc, on top of each other as for "good enough" purposes they will lose.
that said, the reason an intel chip uses so muhc more power than an arm one is not because of the ISA entirely. its because the x86 ISA has so much more functionality. ARM chips are not even 64-bit, they dont have things like mmx / SSE /2/3 super fast floating point devices etc. if ARM cpus were so fast they'd be in servers, and even a 3 year old atom is probably faster than a 3 year old atom.
think about it, if you take a 1GHz athlon XP you probably are still faster than an ARM core right now on integer and fpu compute type things ( i cant substantiate my claims, but i'd think this is probably true). fab a 1ghz athlon XP on the latest 40nm tsmc process with all the power saving measures and clock gating and who knows how efficient it would be.