This seems really interesting, and useful – thanks!
I was curious about any (redacted) details of the kinds of "large systems" for which this would be useful, but the docs in the PR are pretty clear – this is a "dead-simple way to parallelize" dynamic supervisors by running a group of such supervisors, one for each system (CPU) core.