I think the inductor is way too big. Just thinking about this for a moment, at no load the inductor current is at a steady state value. As soon as you apply a load the output voltage drop, the FB voltage drop, the NMOS switch stays on for a longer and longer time. However, the inductor current is going to ramp up slow since the slew rate of the current will be 5V/3mH(ignoring inductor resistance and Ron). It is going to take the inductor many cycles to build up the current to support any type of load at 12V. For instance you are going to get .001666A/uS slew rate with that inductor. If you dropped down to lets say a 100uH inductor that value goes up to 50mA/uS. Even at 15KHz, I think a 100uH should be plenty of inductance(you can probably go even smaller). So now you will see your output voltage dip and it should recover much faster, due to the inductor current ramping faster. I would bet if you let your simulation run out much longer with that large of an inductor you would see the output voltage recover eventually.