I think maybe you are looking for hadoop or ore map reduce based system? Possibly i misunderstood your question.
In this case, my question would be:
How does a map/reduce system redistribute the chunks to balance the load if some chunks are more accessed than others?