I've been thinking about this, and I think the answer is that when a fan is windmilling, the fan blade airfoils are basically working as intended - the angle of attack of the blades is within the range that the flow won't separate and the airfoil stall. Hence the pressure loss through a windmilling fan is small. If the fan is locked, the incoming air is hitting the fan blade at a very high angle of attack - beyond the stall angle - so the fan blades are effectively stalled with the associated large pressure drop (i.e. drag). So there is a greater pressure loss through a locked fan than a windmilling fan, and that equates to more aerodynamic drag. I'm unaware of any engines that drive accessories off of the LP rotor so that doesn't come into play (the core would be a different question).
Vessbot, the helicopter analogy is invalid because of the way the blade pitch is varied during an auto-rotation. If you measured the rate of descent of a chopper with a fixed rotor, you'd find it initially would descend slower than one performing an auto-rotation maneuver. With a fixed rotor and a vertical descent, the air will be hitting the rotor blade nearly perpendicular - making the rotor basically just a big flat plate to the air (the drag coefficient of a flat plate perpendicular to the airflow is nearly 1.0 - which is roughly the same as a parachute. In short fixed rotor blades become a huge drag source during a vertical descent. In contrast, the blade pitch during an auto-rotation is controlled to specifically prevent it from stalling - using the resultant rotational lift component to accelerate the rotors to a high speed so that the kinetic energy an be used to create lift and slow the descent in the last second before impact. There is some induced drag associated with creating the lift that provides the rotational acceleration, but it's much less than the drag created by a fully stalled rotor blade.