> people who build infrastructure and do performance > analysis.
What sort of infrastructure are you referring to? Do you mean the JVM itself? Mathematical libraries? Web frameworks?
I'll go back to my original post, Cliff showed an example of a library getting a 30x speed up using today's tools and just fixing the obvious (once profiled) issues. A 50% (or even 100% or 1000%) slow down in non-hot code doesn't seem like it matters all that much much for most systems, given how fast things are at baseline these days, and how there's almost always a database or disk waiting to murder your performance just down the road.
I'm biased by my work environment, but we work in a a JVM langauge that is, in places, 1000x slower than java (we are fixing that by going to bytecode.) But when we profile, even if our language were infinitely fast, it would only net us a 20% or so increase in end-user speed because other stuff (the database, the network, gc, etc.) is getting in the way. So our semantic gap is orders of magnitude greater than that of regular JVM-based language developers, and yet, for our apps anyway, even a perfect performance model in the head of the language designers wouldn't net us that much.
It might be bothersome to the language designers, but I don't know how much practical effect it has beyond making them grumpy, especially given how fast everything already is.