Several caching techniques have been used to reduce the bandwidth consumption and to provide faster answers in P2P systems. In this paper, we address the problem of reducing unnecessary traffic in the Hybrid Overlay Network (HON), which consists in organizing peers and data in an n-dimensional feature space for efficient similarity search. We propose a distributed caching schema that group similar queries to increase the success hit and avoid redundancy. We show through extensive simulations that caching in HON decreases significantly the query scope improving search performance. General Terms Measurement, Performance Keywords P2P Networks, Similarity Search, Caching