I would agree with the advantage in an e-stop event. If a stepper is spinning at medium to high rate and is stopped hard you will obviously lose a bunch of steps.
But I am not sure if I buy the trajectory theory. Assuming the controller is capable of spitting out the pulses when they are due according to the path geometry, IMO the worst that can happen due to acceleration is that the stepper is lagging by a half step. If it was any more, then the torque maximum is exceeded and you will indeed lose a full or more steps. But then the drive is already under-designed. In a well designed open loop system with reasonable margin, the angular error of the stepper should be well less than a half step. In most routers that would be less than the other mechanical uncertainties of the drive (backlash, linearity, flex etc.).
As for the position memory when turned off, at least the Planet-CNC USB stepper controller that I am using, has that too. When I turn on the machine it will show the last position it had when I turned it off. I can jog the axes without the PC running. This is assuming that nothing moved the axes while turned off (e.g. drooping z-axis).
Box Joint and Dovetail CAM software here: WWW.TAILMAKER.NET