Systems programming in "scripting" languages
tompinckney.com
tompinckney.com
Consider, say, a database. In its equilibrium state, serving a read-biased workload, the DBMS's ability to use every cotton-picking byte of RAM to cache precious blocks will powerfully determine performance. The workload only really becomes "I/O bound" once you run out of RAM, and a runtime that exacts an X% tax on RAM will become I/O bound that much more quickly.
This seems to be an inherent trade-off in the "automatic memory management" paradigm; even Sun's JVM is not immune to this. It's also the right trade-off for a huge majority of code. But it's one of the reasons systems programs in minority languages are driving into a stiff wind.
In fact, in some cases you can get better performance out of a language built for multithreading like Haskell or Lisp than out of anything but the most highly optimized, multithreaded C code.
I can't comment on Haskell, having no knowledge at all of it.
Erlang is rather more in the region of what your comment seems to be getting at, from my perspective.
Perhaps I mispoke, being the very earliest of beginners in Lisp. I had thought that in realms of compiler controlled concurrency for certain types of functional programs it was very similar to Haskell in that respect. I apologize if I erred.