See bug 37396 for some discussion; it would be nice to be able to replay or rerun the results of a previous NRWT run, preserving the sharding, ordering, DRT restarting, and other behavior as much as possible. Right now we don't really store enough information to make that doable, but the cleanup outlined in bug 103824 will fix that. Theoretically once we supported this we could build a simple tool on top of it that would automatically determine N equally-long-running shards for N workers to make the test runs more deterministic as well, which might be worth doing (to reduce flakiness).