This work was supported by Hewlett-Packard and the National Science Foundation CAREER Grant under CCR-96-2386. The authors would like to thank Yasuhiko Miyazaki and Lintian Qiao for their contribution to the implementation of the hierarchical VOD system and measurements of its performance characteristics. The authors would also like to thank the VOSAIC team for supplying logfiles that provided data for the empirical study and Richard Friedrich for useful discussions during the course this work.