* isSpace handles tabs, but looking at a single byte at a time it won't handle all the multibyte space symbols you can have in unicode. If you read further down, they rip out the remains of unicode handling for further speed improvements.

* Looking at a single byte at a time, it presumably only handles the "C" locale :) They don't say what locale GNU wc was tested with (if it's not LANG=C, that benchmark should be re-run)

* --max-line-length? no. But I'm guessing GNU wc isn't benchmarked with that option on (can't find the invocation in the blog post though)

* data State { ws, bs, ls } keeps count of words, bytes (more honest than calling it characters) and lines.