Do you have a firewall between the cells and the vCenter server? We maintain an open connection at all times. I have seen things like this happen when the Firewall has a process to kill idle TCP connections.
One test I have done is to deploy a cell in the same ESXi host as the destination vCenter (assuming it's virtual), and force the proxy over to that cell for testing (stop all other cells). If the error goes away, it's a network fabric problem.
If you have a vcloud-container-debug.log which encapsulates the time frame (and can post it here), I should be able to take a look.