>We have come to understand that developer time is vastly more valuable than computing resources
I get that, but still, 30 servers is pretty expensive. Having a full time engineer just to manage the cluster and write load balancing code is pretty expensive. Having down time because of a usage spike is pretty expensive. I'd rather pay for the smarter dev team myself, but I'm an engineer, so I'm biased.
Though fundamentally I wasn't talking about servers above. I was talking about desktops, and mostly Windows desktops. And having to install Python because that's what one developer prefers, and .NET because of another preference, and Perl because of a third, and Ruby because of a fourth, and Java because of a fifth...where does it end?
My own "scripting language" of choice is Lua. It's TINY, it's faster than all of the above (except, under some circumstances, Java -- but give Mike Pall another year or two and I bet LuaJIT will be beating Java in all the benchmarks instead of just many of them), and using it gives developers that productivity boost you're rightly saying is important.
And if you need more speed than Lua can give you, Go is a good option, if it comes to that. But as you point out, not every app needs every cycle.
[1] http://blog.iron.io/2013/03/how-we-went-from-30-servers-to-2...
We have come to understand that developer time is vastly more valuable than computing resources in the overwhelming majority of applications - the only real disagreement is how many orders of magnitude of inefficiency we consider tolerable. Even the most resource-constrained applications are using higher-level languages and computationally inefficient designs, because it's much easier to spec a faster chip than a smarter dev team.
Most applications aren't getting significantly slower. Web browsers are a perfect example of this; Almost every aspect of their performance has improved vastly over the past few years, but as a result we're building much more sophisticated web apps and doing things in the browser that were once purely the domain of highly optimised native apps. It's a near-perfect example of Jevons Paradox. Improving the efficiency of a process reduces the cost of the output, which leads to increased demand for the inputs to that process.
My current computer has 700,000x more RAM than my first computer. If 99% of that memory is wasted, I'm still doing pretty well.