Your best bet is to pull a detailed CM trace to diagnose this. If it is easier for you, if you can get a packet capture of the SIP traffic during a failure between the two nodes, that could also suffice.
On a hunch, if one side keeps ringing when the other side answers, I'd guess it's either a firewall issue (200 OK or ACK being sent but not making it to the other side) or perhaps something with rel1xx & PRACK.
Is there a CUBE between the two clusters?