Peer-to-peer (P2P) applications such as P2P video streaming and internet video calling have gained tremendous popularity and are expected to be vastly increasing in the next few years. However, low-cost large-scale video services have remained an intangible goal. The ethernet passive optical network (EPON) is being regarded as one of the promising for next-generation optical access solutions in the access networks attempt to tackle this problem but facing a major challenge to offer scalable large-scale video services. Therefore, in this paper, we propose an architecture which combines the advantages of EPON and P2P architecture to provide scalable Internet Protocol delivery multimedia services and improve quality-of-services. In the proposed architecture, we design new optical network unit (ONU) mechanisms, which support traffic redirection communication among ONUs in combination with caching. Thus, it can reduce the resource consumption and add extra downstream bandwidth at the optical line terminal since the intra-PON traffic is not necessary to be buffered and scheduled in the downstream direction. Finally, we propose a “Redirect” dynamic bandwidth allocation scheme, which can support intra-PON traffic redirection and intertraffic bandwidth allocation. Simulation results have shown that our proposed architecture can improve the overall QoS in terms of end-to-end delay, jitter, system throughput, fairness, and packet dropping rate.