A fundamental problem in peer-to-peer (P2P) systems is how to efficiently locate the node that stores a particular data item. As a hot issue, a number of structured P2P lookup protocols have been presented during the past few years. Unfortunately, current protocols either have uneven routing patterns or incur unnecessary high latency hops, and thus lead to inefficient routing. In this paper, a cluster-based and routing balanced P2P protocol, called CRP, is proposed. In this protocol, all participant nodes are grouped into clusters according to cost in latency, so inner-cluster hops have lower average path latencies. Inside a cluster, each node constructs a balanced lookup tree to evenly distribute routing traffic and reduce routing bottlenecks. In routing path, most hops are inner-cluster hops, and thus CRP can lead to efficient routing. A set of simulations to justify CRP is conducted, which shows that CRP is effective in reducing the average path latency and in evenly distributing routing traffic among sibling nodes.