Comment # 1 on bug 1042551 from Andreas Stieger

(In reply to Thomas Wagner from comment #0)
> Using ext4, this setup works well. Using btrfs it works until os-prober is
> launched (IMHO due to kernel or grub update) on the host.
> os-prober tries to mount all btrfs partitions regardless if they are already
> mount in a client. This destroys the btrfs filesystem.

Interesting. 

Looking at /usr/lib/os-probes/50mounted-tests it would seem that btrfs is
handled in a special fashion, in particular not being mounted read-only:

> if [ "$types" = btrfs ]; then
> [...]
> else
> [...]
>                 if mount -o ro -t "$type" "$partition" "$tmpmnt" 2>/dev/null;         

and later:

> # all btrfs subvol processing here.  Handle both unmounted and
> # mounted subvolumes.
> if [ "$types" = btrfs ]; then
>         debug "begin btrfs processing for $UUID"
>         # note that the btrfs volume must not be mounted ro
>         if mount -t btrfs -U "$UUID" "$tmpmnt"  2>/dev/null; then
>                 debug "btrfs volume $UUID mounted"

So the developer must have been aware of a requirements of mounting btrfs
read-write here, which breaks the KVM scenario. Assigning to SLE maintainer.