For one, with a 1 file per row (entity) design, any larger query would require reading from potentially millions of files, which I imagine would be nightmarishly inefficient.