I tried again with the last successful settings on a fresh MOSFET.
Gate threshold voltage measured right during the torture, by adjusting test voltage to the point where I could just see the led blinking. This is crude, hence the graph isn't very accurate. But good enough to see trends.
The total timespan of the plot is 20 hours.
I haven't converted the mosfet into depletion this time, as the process seems to have slowed down too much. I'll try cranking up the torture voltage a bit to see what happens.