logic that expects "compute" cores to have to do I/O whether database or otherwise is quickly going to find all those "parallel" engines ... serialized. Working one-at-a-time or nearly so in a work flow that can actually be considerably slower than having ... one process. You will need to design more infrastructure here than either MCE or MCE::Shared is by itself really intended to provide, to feed the now-truly-
cluster with data without delaying them, and also disposing of the computed data also without delay. While your example is interesting, it simply would not scale.