When it comes to improving Ag's speed, there's not much more to write about. I've used every dirty trick I know: mmap()ed I/O, threads, Boyer-Moore-Horspool strstr(), PCRE-JIT, etc. Now when I profile, Ag spends the majority of its time reading files off disk. On most systems, this is true even if the files have been cached in memory by the OS. The limiting factor is memory bandwidth.

Anyways, I don't really have the time for long analysis posts these days. I'm busy working on something that captures value in addition to creating it. :)