it is the job of the scheduler of the operating system to assign the processes to different cores. both parent and child process should probably be pinned to different physical cores. not sure, whether miller took that into account, though ...
What I have tried in the past is run one pd for audio and another one for GEM stuff, which worked rather well. I wonder if it would make sense to do the same with 2 pd instances doing audio, and exchange audio between them. Maybe I could try that with Jack.
it depends on the scheduling algorithm implemented in jack2 ... depending on the structure of the jack graph, jack2 may try to execute both processes on the sequentially ...
But I think the latency will be doubled, because the buffers of both pd's would add up...?
from my understanding, pd~ uses a pipelining approach, so you will loose one sample block when crossing pd~ boundaries ...
tim