Continued research into distributed agent-based systems and evolving web based technologies are opening up tremendous possibilities for the deployment of large scale and highly extensible and flexible simulation systems. However, the question remains as to how well agent and web based simulation systems can use underlying network services. This is especially true for complex distributed simulations that have multiple participants and real-time data delivery requirements. This paper addresses the issue of interconnection patterns among web-based simulation agents. We first describe a simple agent-oriented architectural extension to an existing distributed simulation system. Next, we present a set of interconnection algorithms between the agents in the system, under the assumption that only end-host multicast is available. We experimentally evaluate our approach under a number of different scenarios. Our results show that by using a proper connection strategy it is possible to get performance comparable to network-layer multicast for agent-based systems that use end-host multicast.