Wow, oddly relevant - I've been working on a RegExp implementation in JavaScript using similar techniques (minus vectorization of course, RIP SIMD.js). There is no efficient way in JavaScript to replace many patterns (e.g. a list of sentence fragments with optional whitespace, capitalization, pluralization, etc.) against a large document where the replacement requires metadata about the subpattern that was matched (e.g. to replace a term with a link to the term's definition).

I wonder if I should switch gears and try to write a native Node module for this instead. This looks really great.