The issue with SuperCollider’s uneven performance in JACK compared to other JACK clients has been brought up before. See this thread:
Perhaps there are benefits to the PipeWire rewrite of the backend that are not only the difference between JACK and PW but other improvements to the audio thread (e.g. calling in or out, locking some resource) that are also beneficial to the performance. Investigating this could lead to improvements to the current JACK backend as well. This makes Luc’s work doubly interesting in my opinion.