Training for real-time NMF (fluid.nmfmatch~)

Does the FluidNMFMatch piano example in SuperCollider clog my CPU because it can’t/will never be able to run 88 FluidNMFMatch’s on a single CPU? Or is it not optimized in some way/compiled correctly?

Max runs the example at 50% cpu (50% in Max, but 100% in Activity Monitor), but I assume this is running across cores. SC is peaking at 250%, but Activity Monitor only shows about 30%.

So, I am wondering if there is some kind of inefficiency here that has to do with compilation or if this is something that I need to hack to run on multiple servers.