Client network goes down after a WAN lost on a multisubnet cluster

I created a 2 nodes Windows 2012 R2 cluster for SQL AlwaysOn + file share witness. Running on Esxi 5.5 update 2. Each node is hosted on a different host.

Each node has two NICs, one for client communications and one for internal cluster communication.
Each node is located on a different site with different subnets, so the cluster group has two different IPs.
Everything was working fine, after I noticed that both clients networks go down if sometihing happen on the WAN link. In order to reproduce the failure, I disconnected the network cable on the client NIC from one of the nodes.
After the failure, both nodes and quorum are ok because there is still communication trough the internal cluster network.

Cluster log is showing the following messages:
State calculation complete without issuing pings
Resetting interface state calculation state
Changing network state for network XXX to 1

I have checked the following:
-Recommended patches according to MS site.
-Cluster validation report is ok.
-NIC configuration and binding order. Type of NIC vmxnet3.
-Vmware support matrix and patches for the current configuration apparently is ok.
-Uninstalled the antivirus software.
-Windows firewall is disabled.
-Best practice analyzer doesn't show anything important.

How the SQL role will be online or failover if both clients networks are down?

Thanks in advance for the help!