Peer-to-Peer systems have shown a great potential on file sharing in recent years, and developing efficient search techniques has become a crucial research problem. Most unstructured peer-to-peer systems with existing distributed routing algorithms can only run blind search, but not global search due to the lack of cache scheme. To address the problem, a novel key clustering algorithm is proposed, which divides the routing space into two layers, AUT layer and HUB layer. Such an algorithm can perform a well-ordered search from a global view. In order to improve scalability, theoretical results from small-world are utilized. In the improved algorithm, a few shortcuts with distant peers are inserted into the routing tables with some probabilities, and the average path length is reduced. The preliminary simulation results show that the key clustering algorithm with shortcuts is efficient and scalable.