Question/suggestion about performances

I see in
that the output flow is made of K*(K-1) streams.

Is there not a possibility to use all users or only some of them as repeaters…?

That is, any user would receive K-1 flows and send 1 flow to server, but send too 1 or more flows to some other users, depending on its own capacities in CPU and net bandwidth.
This would relieve the main server of the equivalent resources allowing to exceed the current practical limit of 35 users… All this would have to be organized by the server as a great conductor…