And if you are in the situation where you are at 99% cpu utilization and those 3 extra cycles are going to make a real difference, you are running your code on the wrong hardware. At that point you need to distribute your work across better servers. It is a simple fact that adding hardware to improve performance is cheaper than rewriting it in another language because Ada or Ruby or Pascal or whatever is too slow. (And yes, I acknowledge that crappy code will run slow no matter what kind of Cray supercomputer you run it on.)

It’s really quite simple: Language choice should be viewed as an economic decision, not a technical one. When a technical strength or weakness of a language is brought up, the question should be “so what?” — which is inherently an economic assessment. Does 5% faster execution translate into noticeably better user satisfaction and retention, or lower costs, or increased sales, or anything else like that? And does it do so more cost-effectively than the alternatives, when everything else is taken into consideration?