Job stuck on Loading Media endless

We have the following problem:
Newly installed server 2016 with BE 20.6 on a decent server.
Locally 8 TB as deduplication storage and as source a vSPhere Cluster 6.7.0
The jobs run for some days without problems (always full backup) on the deduplication storage. But then jobs stop after a few gigabytes. The message is "Load medium" without error message. A snapshot is created on the vSPhere and nothing happens.
In the resource monitor I see massive reading activity in the DeDup storage and no activity on the vSphere server.
I could not find any error messages. Sometimes the job can be aborted and sometimes only a reboot helps.
The job itself is a file server. Agent is installed everywhere and up to date. We back up as a synthetic backup.
Anybody have any ideas?