10%ish performance gain

I couldn't quite believe it, but since the same thing happened when I
ran it a second time I guess I have to. I shaved 34 seconds off the
scalac compile time on my machine (not rebuilding locker) from 6m56s to
6m22s. It was more like 45s on the first run so that might be a
conservative estimate. The faster time includes all the presently under
discussion Number and String equality logic.