I don't agree with that analysis.

That assumes that when driven with PWM the peak current is equal to the stall current, that is simply not the case.

If the value of L/R is much less than the PWM cycle time and the motor is not moving, then it is very nearly the case (where L is the motor inductance and R is its resistance).

It also assumes a motor with negligible inductance, again not the case.

Yes, I was assuming negligible inductance. That is why I prefixed my post with "In the absence of substantial current smoothing due to motor inductance". The effect of the motor inductance will be to reduce the power dissipation; but unless it has sufficient inductance to substantially smooth the current, then the power dissipation will still be much greater in the PWM case. The power dissipation is the average value of I^2 R, and the average value of I^2 is greater than the square of the average value of I, unless I is constant.

Perhaps you have data on the values of L/R for typical motors that you would care to share?