Last night, rx300 rebooted. After a reboot, the disks were detected
in another order (but still, all were detected!), so the cachefile did
no longer match, leading to breaking the import.
Running `zpool import` manually worked, because that will ignore the
cachefile. So, why do we depend on the cache file on boot up? The added
reliability of zfs-import-scan beats the speed of zfs-import-cache in
any way.