A scalable media conferencing system minimizes delay while maintaining a consistent session across all of the distributed media servers handling the conference session. Each of the media servers sends a list of the local top streams and their importance. A global list of top streams is created and disseminated to all of the media servers. Each of the media servers then forwards any local stream that is on the list of global top streams to all of the other media servers. Once a media server receives all of the streams on the list of global top streams, it locally mixes the streams together and outputs the mixed stream to the endpoints that are locally connected to that media server.