Surely someone doing serious research would download the entire corpus and use grep (or equivalent)?

It wouldn't have the same nice interface and searches may take several seconds, but there's only 60,000 books...