Hi Sören,
the problem in your case is that you also have to restart the group object in your workflow.
the group will get the highest returncode of the jobs manged by the group instance. So the group will abort if one or more jobs handled by the group abort.
So to do so:
You use the RESTART_UC_OBJECT as you do.
In the post condition of the group you restart the group object if the group task is ENDED_NOT_OK
The restart of the group automatically will process your job
Regards
Thomas