300ish to 100ish on 1 RCD bank but 2 ohms for tge other.
This doesn't make sense, as the main contribution to the 100 ohms is probably the rod, which is common to both circuit groups. They should be more or less identical, give or take the fraction of an ohm of cable splitting the feed to the two RCDs.
It suggests you have an N/E fault on one group, contrary to your 'readings are fine' and that one loop impedance test was made with one RCD off, isolating the fault. This would fairly well explain the failure to trip and/or reset under some test conditions. What is the aggregate L+N to earth insulation resistance for the installation, or downstream of each of the two RCDs? How confident are you of the IR readings taking in every circuit and appliance? Do the loop impedances still differ if they are taken with all circuits live?
If an explanation can be found for those different readings that is not due to an N-E fault, and the RCD is not faulty itself, then you are probably going to need a leakage clamp meter to do a differential leakage measurement on the L & N going through the RCD with each of the circuits energised, so that you can see what they are contributing to the total.
BTW, I have not seen this thread before. I hate to say it but from the symptoms, it was pretty obvious that there was a fault with the transformer earthing. The phase imbalance thing was a red herring for various reasons, and an Ra of 300 ohms as you achieved with your second spike, would have been sufficient to allow the RCD to function. What was also apparent was that it was quite likely there were two faults, one at the transformer earth grid and one in the house. You might now be on the trail of that second one...