Improve data loading performance

Build on the results in #6 to improve the performance of the heaviest-weight functions in a tiamat pipeline.

How does data loading performance scale with multi-processing?

Can an increased cache by more RAM improve the loading?