Advances in networking and content delivery systems are enabling challenging provisioning scenarios where a growing number of users access video on demand (VoD), possibly while moving among different points of attachment to the Internet and using different access terminals. This calls for middleware capable of supporting personalized VoD access by dynamically activating intermediate nodes between VoD servers and clients. The paper proposes MUMOC, a dynamic and flexible overlay infrastructure for the distributed caching of both VoD prefixes and VoD metadata. To achieve openness and easy interoperability with legacy VoD services, MUMOC adopts standard XML-based formats, based on both Dublin Core and MPEG7, to represent VoD metadata. First experimental results show that MUMOC significantly reduces bandwidth utilization and user perceived delays. In addition, notwithstanding the dynamic building of the overlay network and the application-level approach, the introduced overhead is compatible with the strict requirements imposed by multimedia distribution over the best-effort Internet.