One more minor nit about benchmarking; using the same interpreter instance to run the code that benchmarks the interpreter isn't believable to me. Someone might tell me that behind the scenes these different array variables are using different data structures, but to the observer this is not clear. I'd rather see the interpreter invoked 5 times and run a piece of code than wonder if there is some sharing, collision, some optimization, or some other hijinx going on in the back room, so to speak.

That being said, I love posts like these. Thanks.