This reduces the chance of duplicate blobs, otherwise the tests fail (make the contents of a blob depend on a pseudo-random number instead of the size, sizes may be duplicate).