Well, that's a bummer, but you are made progress narrowing it down. Since you've examined the smoke units six ways from Sunday, take a closer look at the SFC board. See if there's ANYTHING that might be shorting to the chassis or other wires. Obviously, there's a reason for the issue.
I think one of my text tests would be to rig up a wire to simulate a smoke unit with an 8 ohm resistor and a 4.7k resistor to simulate the thermistor. Use a real smoke motor since they check back-EMF of the motor.
Here's what I'm suggesting. If this rig still causes the thermal runaway, then it's not the smoke units, it's something else that is being triggered by the load of the smoke unit. This is a simulated smoke unit, it's how my RCMC test set tests the smoke outputs of the RCMC.