Frequency compensation will be needed-that is for sure. The 2N6051's have a minimum Hfe of 750, so some significant gain is available in the error amp loop for oscillations to occur.
I like the notion of pulling Vref down to ground based on Isense, a transistor in parallel with the Vref diode, (I will probably use a green LED or something, I've got a few zeners in appropriate values but the LED works fine.) and use the output of the Isense error amp to turn on the transistor, shunting current away from the diode and once 3-4mA is shunted off of it, it will shut off and the transistor's collector would become the voltage reference. This would tend to limit the lowest output current available to whatever the output voltage would be when Vref is a Vce when the transistor is at saturation. This may be high enough still to cause a problem-I don't know yet.
I have thought about using one of the .33 ohm emitter ballast resistors on the pass elements as my current sense element-as the current through the resistor rises, the voltage drop across the resistor rises. So measuring the voltage drop across the ballast resistor and comparing it to the potentiometer setting would generate a current "error" that could be applied to the Vref node.
Using the Vref node in this manner I probably should drop the green LED and use a 6v or 9v zener diode, as this would give the current regulator more headroom to pull the current limit around. The again, the more room I give it to move in, the less likely it is to behave the way I expect it to...
I think by implementing the current limiter in this manner, any oscillation that developed would be current limited too, so if some outside influence or a deteriorating component in the frequency compensation network were to cause an oscillation, the current limiter should be able to cut the output current down enough to prevent damage to the pass elements. Maybe.
Going back to frequency compensation, I thought I had a good start on it, with the 10uF cap across the feedback loop to ground. There may need to be a miller compensation cap from the collector of the pass transistors to the positive-input of the op amp, but I will need some help figuring out how to figure that value.