The person you replied to was talking about bandwidth, not CPU load.
Imagine a user that only has enough bandwidth for a single stream. If there was a server integrating all other user streams into one, that user would be able to watch it without much problem.
Unless, the clients are built in such way that they down-scale the resolution to part of the screen it would occupy on target computers (based on number of participants) before sending video data. Is any software doing this currently?
> Even if the server integrates all in bound streams into one, doing so takes a non-negligible amount of additional time per user, surely.
Integration has to happen at one point. Whether a server does it or each user's computer.