The service overlay network (SON) is an effective mean to deploy end-to-end QoS guaranteed content delivery services on the current Internet. We model the content delivery service on a SON as a service delivery tree (SDT). Previous studies have addressed the optimal distribution tree formulation issues. In this paper, we focus on the problem of maximizing the total operation profit of the SON. In , authors introduced a cost model for optimal bandwidth provisioning in SON. In this paper, we extended this concept and propose an alternative approach to maximize the total effective throughput of SDT as well as to minimize the QoS violation penalty of the SON by service replication. We present both centralized and distributed algorithms for the placement of replicated servers on the SDT. Experiments are carried to quantify the merit, effectiveness and the scalability of the proposed service replication algorithm. In particular, the performance gain is very close to the exhaustive search. The algorithm performs well when we scale up the service overlay networks. Finally, we show that one only needs to perform a small number of replications to attain the optimal gain.