Well, no-one replied to this, but it worked. After numerous tests we had the transfer down to a tee. All that (we thought
) was required was a small change in the udev config, since the new hardware has different network card drivers.
We did the 'live' client restore tonight. Everything was fine until booting with the new hardware. It turns out someone had turned on a restore-from-swap package on the client's system (which we hadn't planned for). This caused the boot to fail, since there was no restore image in the hot-swapped swap partition. (If you follow
). Anyway, some frantic googling turned up that this is quite a common problem, but didn't really provide a fix. To avoid the restore-from-swap, you can boot with a 'noresume' kernel option, but by the time we'd figured that out, the RAID arrays were toast (although filesystem checks on the underlying partitions were clean).
Booting into knoppix and manually rebuilding the RAID with only one device (the hot-swapped drive's partitions in each case), then rebooting fixed it for us, and everything is working as it should.
While the total downtime was in excess of an hour (mostly while we scratched our heads wondering if we'd killed everything), this was still a lot better than the 5+ hours it would take to transfer over the network, plus altering the config files and everything else, with the added bonus that if it came to it we could stick the old machine back in unchanged.
All in all, RAID is great, and we know what to do next time.