Reliable multicast is useful for replication and in support of publish-subscribe notification. However, many of the most interesting applications give rise to huge numbers of multicast groups with heavily overlapping sets of receivers, large groups, or high rates of dynamism. Existing multicast systems scale poorly in one or more of these respects. This paper describes QuickSilver Scalable Multicast (QSM), a platform exhibiting significantly improved scalability. Key advances involve new ways of handling time and scheduling, adaptive response to observed traffic patterns, and better handling of disturbances.