The docs now recommend a thermal shutdown switch on the heatsink. I mistakenly assumed that the diodes would behave as they do in a non-darlington output stage. I did find a thermal shutdown circuit consisting of a couple of transistors and a zener diode, which I'll try to figure out values for sometime. Not sure how to go about testing such a thing, though.
I've also added a diode clamp to the input to prevent overloading. (Obviously this will add distortion if the input goes past about a volt, but also limits the power output to a safe level.)
My goal was an easy build, not great quality. Hence the diode biasing (less to adjust) and output capacitor, etc. Thanks for the insights.