While multicast has been studied extensively in many domains such as content streaming and file sharing, there is little research applying it to synchronous collaborations involving shared access to a distributed object. Based on several cases of real-world collaborations involving instant messaging, distributed lectures, and computationally-intensive collaborative game playing, we show that compared to traditional centralized and replicated collaboration architectures, a new bi-architecture collaboration system model with multicasting support can improve response, feedthrough, and task completion times. In addition, we show that to optimize performance, the set of traditionally considered factors, consisting of network delays and transmission costs, must be expanded to include several new factors, such as processing costs, scheduling policies, and think times. In one or more of the real-world collaborations we consider, we show that multicast (a) can increase feedthrough times if processing costs and scheduling policies are not considered and (b) may degrade or improve task completion times depending on the cost of computing the multicast overlay.