I think I was able to answer my own question...which apparently is Yes, as long as you don't need to power on more VM's than the physical resources can support.
I tested this by making only one GPU available to assign as a "Shared PCI Device". I edited two desktop vm to use the grid_k280q profile (4096MB frame buffer). I was then only able to power on one VM. Attempt to power on the second VM resulted int this error:
