Since massive objects are stored in the storage systems, and may be scattered, this situation increases the search time to access the objects. However, traditional VRML system never considers the problem of how to reduce access times of objects in the storage systems. Meanwhile, clustering methodology is particularly appropriate for the exploration of interrelationships among objects to cut down the access time. Besides, we introduce the relationship measures among traversal paths, views and objects. Based on these relationship measures, the clustering algorithm will determine how to cluster and the optimal physical organization of those VRML objects on disks. In addition, we suggest two clustering criteria - intra-pattern similarity matrix and inter-pattern distance table. Our experimental evaluation on the VRML data set shows that our algorithm doesn't only significantly cut down the access time, but also enhance the accuracy of data prefetch.