It definitely seems obvious after running an

explain

on an org with 20k+ members when you're wondering why performance is tanking 😅

In that regard, we've recently added a new parameter that allows you to tweak the number of items on a dispatched request. So if you have a wide relationship like 20K, today you'd have to dispatch 200 requests because the default is 100 elements per request. You can now increase this with the new

dispatch-chunk-size

flag, which should make those wide rels faster to evaluate. Also the new

LookupResources2

experiment should make LR calls faster too.