I mentioned it when it came up on another thread. Compare apples and apples -- use one of the standard corpuses when running bench marks.

Ian Witten put together the Calgary corpus - https://en.m.wikipedia.org/wiki/Calgary_corpus