An optimized caching mechanism for byte range requests from a web service is provided. When a large file that is not already in cache or being transferred is requested, a background thread may be created to transfer the file and a shared data structure created to track portions of the file that are transferred. For each subsequent request for portions of the same file, the data may be sent back in chunks and the request to read each chunk from the file blocked until that chunk is filled by the background thread. Thus, the locally stored and partially filled file is shared among multiple requestors.