Partitioning data into 80 bounded stores achieves linear performance at 2.6 ms/file, which is 3.1× faster overall than a single monolithic store.