The approach I've tested so far is boiled down into this (rather contrived) example: https://gist.github.com/wyager/df1809badc7c6a75cd5f

Without optimization, each add1 adds about 0.37 seconds. With optimization, each add1 adds about 0.16 seconds. That's over twice as fast! Of course, this is very much a "lab environment". 

I suspect you might be able to make the rewrite rules better with a few tricks, but I haven't gotten around to testing them yet.

--Will