Changing the number of voices in a poly seems to free the existing patchers. I'm kind of flustered because I thought this was working perfectly but realized the fatal flaw in my strategy.
Am I just going to have to figure out another technique? With jit.gl.model it would work to create the maximum number of voices only to load the models as necessary (though this makes the patch a bit sluggish), but with jit.gl.videoplane I would need to have to load them all and disable them as necessary. Not ideal by any means. Another possibility I suppose is re-reading the models when new voices are added, but this gets cumbersome and makes the whole thing less scalable.