Comment by Zizzle

It also seems like a task with inherent parallelism.

Upload the corpus and throw a bunch of cloud instances at it.