Another interesting way to handle this would be to signal the RAID system with TRIM (or UNMAP on SCSI systems). The RAID system could keep running at the block level with no filesystem knowledge, but still gain the benefits of knowing exactly what would need to be rebuilt in a failure.