Regarding running on the GPU: I did have a branch which ran the velocity and constraint solvers on the GPU using OpenCL, but the overhead of memory bandwidth negated most of the performance gains. And on top of that, neither console runs OpenCL so it would have been a PC only solution. Most importantly though, the bottleneck is not in either of the solvers! It’s in the rigid body collision phase… The solvers can handle more than 120k particles at 60fps using the Parallel build of VICODynamics. This PhysX bottleneck is my biggest performance issue. I’m not sure nVidia would give any insight on this issue 
.
Regarding varying the number of particles used to simulate a mesh: Most definitely a yes! This will be implemented, most likely in Feature Update 3.
Happy to hear you enjoy the plugin! Please feel free to post any work you’ve used the plugin in, I’d love to see it.