In this paper, we present an approach, named RS2D, to retrieve top most semantic relevant OWL-S services in unstructured P2P networks with minimal communication overhead. Using RS2D, each peer determines the average query response behaviour of its immediate neighbors and routes the query for semantic relevant services to those with minimal naive bayesian risk of failure in terms of both communication and semantic gain. Experimental evaluation of the performance of integrated risk based service retrieval is compared with that of other relevant P2P service retrieval systems.