preferably no, probably yes. We'd like to hold off on gpu->cpu downloads for now as undelayed downloads cause full pipeline flushes where the cpu has to wait for the gpu to complete and then the gpu having to wait for the cpu to provide more work resulting in the machine doing a whole lot of nothing for a while. We'd need undelayed downloads when they're done mid-frame for a result that makes sense logically.
We could do delayed downloads but it needs to be obvious that this is happening so that it's clear that it can be used for driving some graphics or w/e but not for logic. Additionally we might provide compute support which could enable some evaluations to happen on the gpu, outputting their results into gpu buffers which can then be used by other gpu based nodes.
In the end if you want to send a pixel color out over osc or something you need to have a download so we cant get around that, but yeah for the other cases it's probably better to have the tools of choice available before providing the clown shoes enabling you to stumble over your own feet.