Gladly we could fix the problem. We had to set the following config setting (because our servers each have a management + productive IP):
"com.mattermost.calls": {
(...)
"tcpserveraddress": "<productive IP>",
"udpserveraddress": "<productive IP>",
(...)
},