If you continue on this way you'll do something really wrong and you'll lose all your data; don't say after that I didn't warned you before...
Actually I don't think you've warned me. You've said this is bad without much of an explanation. You've suggested a single stripe disk which is a single point of failure, so that's clearly not the way to go.
so what is bad? - the use of a file as a device in a zpool? Well I won't use this as I had thought that I needed 4 disks for a RAIDZ2 zpool when in fact I can get away with a 3 disk Z2 pool. It certainly better not to use a degraded pool for copying data. Obviously the capacity is lower but thankfully in this case it is enough. It seems like I've made the decision to migrate just in time.
I'm doing the migration using zfs send/receive and I'm keeping snapshots (it seems to me to be outright dangerous to just throw away your snapshots). There will obviously be points where I have to degrade the pools but that's just what happens when you are swapping disks and you don't have enough.
What I'm considering is to take a single disk out of the Z1 pool - use that as a single disk and copy everything from the Z1 pool to that. Then take the 2 remaining disks in the Z1 pool and add them to the Z2 pool. Finally removing the 3Tb disk (2 partitions) and releasing that for it's intended use.
thanks for the feedback. Even when I don't agree, it helps me consider all the possibilities.