Ah - I hadn't looked at the actual processing code. As that code is mostly linear CSV processing and some counting, I concur with your analysis that there is little to be gained from changing the processing, as it most likely is IO.