Hi,
IMHO this is not a particularly "difficult" problem, but several of the important details mentioned in this thread are being overlooked. Firstly. Joe has linked to the motor data which specifies the voltage range is 6 - 24 volts (nominal 12), the current at maximum efficiency is 110 mA and the stall current is 420 mA (presumably at 12 volts).
As premelec has said, to detect the stall you must measure the current, most simply by putting a resistor in series with the Source pin of the FET (so the voltage drop can be measured relative to earth). One Ohm is a very convenient value and probably optimum for this application. Half an amp (d.c.) dissipates a quarter of a watt in one ohm, which is the typical rating of a standard CR/MR25 resistor.
The PICaxe ADC input pin could measure the voltage directly across this resistor, but it would be better to connect via a resistor of 1k - 10k. This will protect the PICaxe against any "accidental" current flow and gives the option to add a capacitor (perhaps 100 nF) across the ADC input, to act as a low-pass-filter (to average the value of the current).
But as goom has indicated, the stall current will vary with the PWM ratio in a rather complex way. Firstly, the PWM reduces the effective voltage applied to the motor, so the stall current will be less. Secondly, the resistor is measuring the motor current only when the FET is switched ON*, but the motor current continues to flow (via the catching diode to the supply rail) when the FET is OFF. Normally, the inductance of the motor will filter the current to almost d.c. (with some ripple), unless the PWM frequency is very low**.
* A resistor directly in series with the motor could be better, but this would require measuring the voltage drop relative to the (12 volt) supply rail, or "inverting" the whole output stage (motor to ground) which would need a high side switch (e.g. P-channel FET and driver).
** Very low frequency PWM is sometimes used with small model motors because it offers subjectively higher torque, by giving "kicks" that overcome the static friction (rather as a hammer is more effective at fixing a nail than trying to just push on it). But the functional analysis is likely to be more complex.
The structure of the software for stall detection is mainly a matter of personal choice. The voltage drop across 1 ohm won't be high, but a READADC10 should give sufficient resolution, or the sensitivity could be increased by using a FVR for the ADC reference. A capacitor across the ADC input might help filter a (single) ADC reading, but taking a burst of multiple ADC readings may be more informative. They could be simply averaged, or a peak value extracted. Then the value will need to be scaled or compensated to take into account the present PWM value.
Finally, it hasn't been mentioned that the stall current is very easy to measure or calculate. Just Ohms Law: The stall current is given simply by the (effective) voltage applied to the motor divided by its dc resistance.
Cheers, Alan.