In the Metro Last Light benchmark using a GTX 780 dedicated to PhysX w/a GTX 980Ti and then a 980Ti dedicated to PhysX w/my 1080Ti it never made much of a difference. When I dedicated the 980Ti to PhysX along w/the 1080Ti I saw a 14% peak load and 169 MiB of VRAM utilization on the 980Ti. Maybe Batman: Arkham City or Borderlands 2 or Fallout 4 or Killing Floor 2 might see more of a difference. Someone here did say he saw a diff w/a dedicated physX GPU in Fallout 4. I never really bothered to do any more testing after seeing the disappointing results w/Metro Last Light. The interesting thing is that if you set PhysX to auto you get slightly different results than if you set PhysX to GPU.
If you can it would be great if you could do some testing on your own. I'd suggest two benchmarks w/one of your 580's installed:
1. PhysX set to CPU
2. PhysX set to GPU