Quoting Brad Campbell (lists2009@???):

> Actually, this exact reason is why I moved from Lilo to Grub a few
> moons ago.
>
> It happens *when* one of the primary OS drives dies in your server
> and you get a reboot before you have a chance to fix the array.
>
> Example (because this is what triggered the move for me). You have a
> RAID1 with your root and boot on it. LILO is installed on those
> disks, and you can tell the BIOS to boot from either and it works
> just fine. One of those drives goes titsup and before you've had a
> chance to deal with it (because it's in an office 13,000km away) you
> get a power outage that exceeds the runtime of the UPS. The system
> cleanly shuts down and never comes back up because LILO does not
> cope with the fact the BIOS has re-ordered the drives.

What a pity you never read the tip in the Boot+Root+Raid+LILO HOWTO that
I referred to upthread:

# BIOS=line -- if your bios is smart enough (most are not) to detect
that that the first disk is missing or failed and will automatically
boot from the second disk, then bios=81 would be the appropriate
entry here. This is more common with SCSI bios than IDE bios. I
simply plan on relocating the drive so it will replace the dead
drive C: in the event of failure of the primary boot drive.

http://www.tldp.org/HOWTO/Boot+Root+Raid+LILO-3.html

That _seems_ to address the exact situation you speak of. I haven't
tested this, but I might eventually get around to doing a test.
(My systems currently use GRUB 0.9x, and my present intention is to
migrate not back to LILO but rather to extlinux.)