Distributed Multimedia Applications (DMA), such as video conferencing and video-on-demand, require dynamic Quality-of-service (QoS) guarantee from end-user machines for their continuous multimedia streams. The users generally start their sessions or chains of tasks at their machines. These random accepted sessions can cause some computers to become highly loaded creating resource bottlenecks while others are lightly loaded or idle. Load balancing can be achieved by minimizing load on heavily loaded/bottleneck machines, i.e. minimizing Bottleneck Resource Utilization (BRU). The paper presents a new multiple-server-client (cluster-server-client) architecture along with a new algorithm Reduced-bru-Multiple-Server(REMS) for video conferencing applications. Exhaustive simulation results show that in terms of minimum relative BRU REMS outperforms than RED and SQ algorithms.