[anthony_r]

> I'm talking about having a compiler which could produce more optimal code in any given case, or most given cases

Funny coincidence, that's exactly what the vast majority of compiler code is for, with tens (or hundreds) of thousands of man-years of work spent on that.

[samhw]

My point, again, is about having a compiler that can produce more optimal code than existing compilers (and whether there's a theoretical upper bound preventing them from more closely approaching the handwritten assembly in this case).

I'm not quite sure what else you could have thought the word 'more' was referring to?