Be interesting to have a benchmark on how much quicker/slower it is working with data in different chunking schemes and file layouts and schemes.