In modern information systems the dominant retrieval paradigms have shifted from exact matching towards retrieving a list of the most relevant objects. This is because users usually are not satisfied by arbitrarily large and unordered answer sets, but rather interested in small sets of ”best” answers. All popular ranking approaches need collection-wide information. This poses a challenge in the P2P context where each peer has only local knowledge. We discuss how to deal with collection-wide information in P2P networks and how to optimize query distribution based on query result analysis, and present evaluation results.