This allows the filesystem blobstore to mimic real blobstores more accurately. This also doubles the speed of getBlob by storing the Content-MD5 header instead of recalculating it.