Scheme benchmarking with a meta-circular interpreter

This benchmark is missing Chicken Scheme: http://www.call-cc.org/

No Gambit-C? That's why Racket is a winner.)

There is also Bigloo and MIT Scheme (which is able to produce native code).

I wonder how much of Racket's improvement over time really has to do with underlying changed to GNU Lightening (the library that underpins its JITing). Also, the interpreter and benchmarks look like they inherently single-threaded, so even though the server hardware looks beefy compared to the MacBook, it's really only using one of thoses (probably slower) cores. There are Scheme and CL implementations that make it easier to write code that parallelizes easier than is possible in Racket -- so anything beyond simple apples-to-apples code like this benchmark would show very different results.

> As we know, in comparison to compilers, interpreters have “interpretive overhead” when they execute programs. What if you run the interpreter itself inside a copy of itself? The overhead roughly multiplies.

Wouldn't benchmarking this factor be more interesting than just benchmarking multiple compilers for a a fixed n levels of interpretation-recursion?

Faster than Clozure Common Lisp? That just itches my skeptic button.