You do realize you would need a different translation layer for each file system type including the RAW ones used by database engines and the OS would have to know how to update this translation layer and the performance gain would probably be negative due to the extra lookups.