Decentralized search by routing queries over a network is fast emerging as an important research problem, with potential applications in social search as well as peer-to-peer networks. In this paper, we introduce a novel Social Query Model (SQM) for decentralized search, which factors in realistic elements such as expertise levels and response rates of nodes, and has the Pagerank model and certain Markov Decision Processes as special cases. In the context of the model, we establish the existence of a query routing policy that is simultaneously optimal for all nodes, in that no subset of nodes will jointly have any incentive to use a different local routing policy. For computing the optimal policy, we present an efficient distributed approximation algorithm that is almost linear in the number of edges in the network. Extensive experiments on both simulated random graphs and real small-world networks demonstrate the potential of our model and the effectiveness of the proposed routing algorithm.