A set of program elements (e.g., transcoders) are grouped together as an administrative unit. Instead of caching the individual outputs of each program element, preferably only the aggregate output of the set of program elements, taken as a whole, is cached. The inventive technique enables the effective re-use of intermediate content. In an illustrative client-server based implementation involving a transcoding service located at a server, the cached information may be shared across multiple server instances to obviate redundant processing. With the present invention, a caching mechanism in a complex software system may be extended in a user-configurable manner by setting up optimal intermediate caching points that are defined by groups of programs used in long computations.