These are always interesting to read because you get to see runtime quirks. I'm surprised there was so much function call overhead, for example. And it's interesting you can bypass range checkong.

The most important thing, though, is the process: measure then optimize.