My gut reaction is that you’d want to lever trame’s client server to render with C++ on the server side. If you are rendering huge point clouds, I expect most client-side browsers would choke. But I am far from knowledgeable about this area
My worry is that if you resample based on camera position. You will keep sending different set of points every time. To keep that reactive (network wise), you will have to use a reasonable number of points as no caching will happen on the client side (always a different set of points).
I mean you could achieve that by sending a new set of point when you stop interacting. That way you might get a reasonable behavior assuming the network latency for exchanging that sampled point cloud is not too big.