When openmediavault-luksencryption is applied to a device like /dev/md0, it uses the whole device instead of a partition that covers the whole device.
I was so unlucky as to have the automatic partition scanning of Linux to think that there existed two partitions /dev/md0p1 and /dev/md0p2 on my device with 1.5 TByte and 750 GByte each. I think this was because of a random combination of data in the LUKS header or in the content itself. When I tested via parted, there was no valid label on the device.
This led to many problems:
1. In the GUI RAID overview, those devices were shown. After I verified that they devices were unneeded, I tried to remove them in the GUI, thereby removing /dev/md0 as well. Probably the mere existence of these devices would have destroyed the drive content anyway as the recovery of those devices had not begun yet.
2. Upon a reboot, when the RAID5 buildup was not finished yet, the rebuild did not continue, saying "resync=PENDING".
Since I think that one cannot disable probing for partitions, the proper way of handling this would be to create a GPT disk label and a large partition covering the whole device and put the LUKS containers into that partition. When I tried this manually, I could not choose the partition for encryption via the GUI.
You cannot rely on "random data" on a device not to trigger this behaviour.