And because the file output will almost certainly be the limiting step, am I being silly in parallelizing this in the first place?

No one can give you a good answer, without doing a really good benchmark, but there are two things you ought to consider.

• You're doing IO. That's a lot slower than performing quite a few calculations. In other words, you have to be able to save an impressively large number of computations to make up for performing IO.
• If you have one process aggregating all of the results of multiple other processes, you have a point of synchronization and that implies some sort of ordering and slowdown. This may eat up a lot of the other benefit of parallelization.

As far as possible, routines that benefit from parallelization should be completely independent. If you really need parallel processing here, have each process write to its own file, then merge them (if necessary).