Right now we're having issues when the number of files in a directory is very large. On the Grid, operations on directories with more than ~50,000 files are very slow, and directories literally can't contain more than 1,000,000 files. We're hitting this limit for some computations.

I suggest we separate the various files by homological degree.