Do you use anything like a trigram index (see rsc's wonderful posts about how Google Code Search worked, and https://code.google.com/p/codesearch/ for a Go implementation) to speed up the regex codepath?
I'm using a different data structure -- a suffix array -- but the concept is pretty similar. I started work on this before Russ released his codesearch implementation, but I did read his blog posts while I was working on this.