Are you still operating on a single stream of characters? I was wondering about something more radical, like you start in the middle of the stream, and one thread parses the forward to the end, and the other parses backwards towards the beginning.
That's an option for a potential 2x boost, which I suppose could be good enough in certain contexts, but that's it. In particular, if you have a parser system able to utilize SIMD for all parsing, that'll get arbitrary parallelizability for "free" (at the cost of some log(n) increase in total operations across all threads).
