One possibility I can think of is that the throttle shaft is bent, meaning an imbalance in the air supply to the two banks at idle, but, now I think about it, I think that one's blown out of the water by the manifold multiram valve being open at idle, so that would balance the airflow anyway.
A big enough air leak into one side of the plenum might do it...
What are the fuel trims and lambdas doing on the two banks immediately after they become active, before the misfire, and after a reset of the fuel trims? If they are dragging the fuelling far enough out to cause misfires, then the lambda sensor sill report silly values because the misfire admits air into the exhaust.
If I've understood, it ran correctly for a minute or two, indicating that it probably all went wrong when it went closed-loop, meaning likely a lambda sensor problem. When the lambdas were disconnected, were the fuel trims also reset? It could be that it was running with a stored fuel trim value even though the lambdas are now out of the equation.
A reset of the stored fuel trims and then a datalog of all parameters form cold until the misfire ensues would be interesting.