What's recommended in this case? My guess is that the hardware mirroring device flushes immediately on write, so this should be safe to ignore.But it may not flush the underlying disk's cache, so it may still be a problem.
It just occurred to me that the actual hardware mirroring device is SATA. Does SATA include a similar cache flushing command? If so, then the issue might be asking Acard (who make the SCSI to SATA adapter) to implement the command.
I use the same type of Acard SCSI-IDE adapters in many of my NetBSD machines, so I suppose I might start seeing this issue on them once I move them to NetBSD 5 and enable WAPBL. We'll see.
Thanks, John