Wide-area application level multicasting (ALM) has been an active research area lately and practical algorithms have been derived to deliver QoS results for small to medium group size, covering a good range of cases when such requirements are warranted. These algorithms all assume that the resulting tree is comprised of the members in the session only. However, in a large and collaborative environment, active sessions are likely to consume only a fraction of the total resources. Moreover, some nodes with large capacity can contribute to multiple sessions. While it is intuitive to explore such spare resources, the challenges are 1) how to organize all the available resources into a resource pool and 2) how to discover and subsequently utilize the spare resources to benefit the active sessions. In this paper, we describe SOMO ( Self-Organizing Metadata Overlay ) which is an in-system monitoring service that effectively creates an illusion of a single resource pool made up by machines organized using P2P technologies. Using SOMO, we show practical solutions utilizing spare resources can substantially optimize active ALM sessions. Furthermore, sessions with different priorities occupy resources accordingly. All these are achieved using a hybrid model that combines in-time global knowledge and individual competition without the need of central coordination.