Retrieving resources in a distributed environment is more difficult than finding data in centralised databases. In the last decade P2P system arise as new and effective distributed architectures for resource sharing, but searching in such environments could be difficult and time-consuming. In this paper we discuss efficiency of resource discovery in PROSA, a self organising P2P system heavily inspired by social networks. All routing choices in PROSA are made locally, looking only at the relevance of the next peer to each query. We show that PROSA is able to effectively answer queries for rare documents, forwarding them through the most convenient path to nodes that much probably share matching resources. This result is heavily related to the small-world structure that naturally emerges in PROSA.