Scanner is far slower than it should be

[stevehalliwell]

In some of our benches scanning/tokenisation is far slower than the actual bytecode generation. While not an issue at present it feels wrong.