Rebuilt the lot of it. Done. Worked.
Put it through its paces, wide open, stalled, brake.
Going from wide open, to stalled, to wide open caused the forward mosfets I tested to fail shorted.
However, I did something wrong which ultimately invalidates my test. I never hooked up the return path for the forward schottky diodes. As such the only return path for the 'power off' spikes went through the body diodes instead and they are not really fast enough to handle them.
I was debating if I should hook that return path up to the charging input on my BMS, power output on the BMS, or directly to battery - and bypass the BMS. I never decided.
Time to pack up here so I'll have to replace those mosfets in a day or two.
<EDIT>I believe my testing methodology itself was the source of my eating mosfets here!
If I'm right, my BMS is triggering when I stall the motors, and my test used the relative ground of the battery and not the BMS for the gate voltage.
So.. my gate voltage would go from a comfy 3.8v to a fatal -35v... if/when the BMS starts tripping as the BMS actually interrupts the ground when it trips.
The BMS isn't supposed to trip until 100A, but that is another issue.
Thanks to @Bharbour for his suggestions on the issues I've been having here. It's hard to help without a schematic or pretty much any measurements, but he gave it a shot.. :)