Well, I was wrong. It is not odd behaviour at all :-)
My research on the web indicates that the system will fail miserably. If you were booting from it, if might not even boot.
In the case of failure, the mirror has to be "broken" (i.e. undo the RAID configuration) or "fixed" (i.e. a second drive installed to replace the missing on.)
When a member of a mirrored volume is orphaned, you need to break the mirrored volume to expose the remaining volume as a separate volume. The remaining member of the mirrored volume receives the drive letter that was assigned to the complete mirrored volume. The orphaned volume receives the next available drive letter or a new letter assigned to it.
I think fundamentally that I need to break the mirror
Normally, if a disk in your RAID array fails, you should be able to shut down the server, remove the failing drive, and replace it with an identical (at least in capacity) drive, and start back up, whereupon the system's RAID controller should begin rebuilding itself.
You had problems before with the hardware-based RAID. Well, everything has its advantages and disadvantages. For my uses, OS-dependence is worse than hardware dependence, but it would depend.
My web research also keeps me away from using "dynamic" disks. They have advantages and disadvantages also. Too many users reported slow-downs. I assume it is somehow due to the redundancy (safety) it has, and the way it stores that redundancy.
Manual mirroring, then disabling the disk device itself might be one way to keep a data disk more secure, in the event of viruses. RAID-1 is for quick recovery during a hardware failure, but mirroring error and damage and deletion makes it less useful for all other soft and user problems. A (device) disabled backup with 90% of the data on it, might still have 90% of the data on it, if a virus wipes out everything.