I don't really see that point of the implication that Android Linux isn't optimized everywhere. It isn't, nothing is, there's no point really.
The question isn't when will Intel create a power efficient CPU (as you could argue that they have - such as the Atom Z series) but what will it take them to match ARM's performance per watt. In other words once we have a 1W Atom - which is probably pretty close to the consumption of the A8. What did Intel give up to close that gap? Die shrink? Die size? (aka drop functionality) Or will some other factor make it all moot? (battery tech, market change)
So far you haven't made much of an argument for ARM being more incompatible than various versions of x86. - ie. I can't run AES instructions on anything but the 2nd Gen Nehelem. Not only that (and this may be true for ARM but I don't code in that much) but there are lots of contradictory optimizations in x86. For example just taking block moves what is optimal for 286 (Unrolled loop if move can fit in cache or REP MOV), is suboptimal for 386-486 (unrolled loop for in cache move otherwise REP MOVS - for blocks aligned on word or double word boundaries) which is suboptimal for Pentium (MMX), etc.. this isn't even counting dealing with the wide variety of cache sizes.