We propose a scalable multicast architecture for potentially large overlay networks. Our techniques address suboptimality of the adaptive hybrid error correction (AHEC) scheme in the multicast scenarios. A hierarchical multi-stage multicast tree topology is constructed in order to improve performance of AHEC and guarantee QoS for the multicast clients. The multicast tree is divided into subtrees, called regions. Every region is assigned a control node, which serves the individual redundancy and retransmission requirements of the receivers within the region. Region sizes are bounded by the maximum cost per region, which defines the ability of the control nodes to serve the receivers of the assigned regions. We show that the multistage multicast architecture significantly reduces the amount of redundancy information introduced into the network and brings it closer to the Shannon bound.