To hang or not to hang, this is the question.
And my worst nightmare too.
Recently, I have tested an 89C52 based board using a contactor and a highly inductive load. (A MAX813L was also in the board).
Using a suitable SW in the board I have detected different effects:

1) The MCU enters some loop. The 813L resets the micro.

2) The micro jumps to execute from external memory (which does not exist) an after reaching the FFFF adrress resumes from 0000, but no HW reset is performed.

3) The micro enters a blocking state (maybe latching state) with a high current drain from the power supply. The voltage regulator (7805) limits the current, so the voltage falls down to about 2 volts. The micro does not respond to an external reset. This is really a critical situation.

After some trial and error, I was able to eliminate the third case changing the PCB path of a ground line where 2 transzorbs were connected (RS485 line). This also reduced the other effects.

Though maybe the question is not what the MCU does when fails, but how to eliminate (or reduce) this posibility. Many excellent answers have been already posted on this subject.

Alfredo del Rio.