Access to compound data over a wide-area network is optimized by analyzing metadata within compound data to identify internal and external data streams to be prefetched. Upon receiving or intercepting a network packet including an access request for a data resource, metadata in this data resource is analyzed to identify associated data streams and their storage locations within and/or outside of the data resource. Data streams may be proactively or reactively prefetched. Proactive prefetching identifies and retrieves data streams or portions thereof likely to be accessed by a client based on attributes associated with the data resource. Reactive prefetching identifies portions of data streams associated with received access requests and retrieves additional portions of these data streams. Prefetched data streams or portions thereof are stored in a data storage on the same local network or near to the local network including the client.