Because the speed increase is - on modern, properly tuned filesystems - surprisingly small, due to how RDBMS's manage their pool; by working on large container files, they avoid most of the filesystem overhead.