Under the present invention, a history of requests for data objects are tracked and maintained in a cache log. Based on the history, certain data objects are prefetched into a cache. When a request for a cached data object is later received, the requested data object can be retrieved from the cache and served to the requesting user. Thus, the latency involved with obtaining the data objects from the appropriate sources is eliminated. Further, under the present invention, discard and refresh rules are assigned to each data object on a class basis. Accordingly, data objects in the cache can be refreshed and/or discarded so that the caching operation can be optimized.