Peer-to-Peer (p2p) networks are used by millions for searching content. Recently, clustering algorithms were shown to be useful for helping users find content in such networks. However, p2p networks often exhibit power-law node degree distribution, causing biased results when clustered using current algorithms. In order to overcome this bias, an efficient clustering algorithm is presented, which targets a relaxed optimization of a minimal distance distribution of each cluster with an additional size balancing scheme. Using song similarity graph collected from crawling 1.2 millions users in the Gnutella p2p network, we present methods for improving the ability to search for content and build novel recommendation systems.