I've been suspecting a certain bug for a while. I finally got around to confirming it yesterday evening.
Neither the Windmill's custom autopilots nor Attitude MFD are firing the rear RCS thrusters for pitch moments.
After busting my brain, tearing apart code and staying up way too late trying to figure out why the heck the rear thrusters weren't firing, I finally had to force myself to hang it up and go to bed. I figured it out today, though.
Both the Windmill autopilot and Attitude MFD will set the thruster levels for a roll moment and some time after the pitch force is set. If no roll moment is needed, the thruster levels for the roll thrusters are set to zero. Since the rear thrusters are +/-Y "overloaded" for both roll and pitch (the front ones hypothetically are too, but in actuality, they use different thrusters as defined in the code), the roll code cancels out the setting for those thrusters that was just made by the pitch code. Most Orbiter vessels have separate roll and pitch thrusters, so this isn't a problem for them.
I can think of about four or so solutions for this problem (in addition to just leaving it alone and having under-powered autopilots that impart a bit of translational thrust as well as rotional thrust). Actually, the first two are fixes to the autopilot code, which wouldn't help with third-party autopilots such as Attitude MFD, so I'll go ahead and axe those... (Unless I wanted to overwrite SetAttitudeRotLevel and SetThrusterGroupLevel for additive effects rather than... Naah-- I think the methods defined in the VESSEL class would get called by third party applications anyway, since they're not virtual methods.)
3. Replace all of the thrusters with pairs of logical thrusters that add up the forces from each of the "hypothetical" thrusters. (I keep wanting to say "real" thrusters, but there aren't any real thrusters... It's all a computer simulation.) the overlapping exaust textures probably wouldn't be too noticeable, but it'd take a bit of fancy code to redefine the logical thrusters each time one of the hypothetical thrusters is armed or disarmed.
4. Add more logical thrusters for pitch moments.
Oh...
5. Use just the front thrusters for rolling-- though that'd look a bit unrealistic, as it'd put extra stress on the truss.
Hmm... I think #4 might be the best option. Any opinions?