I realise I have never thanked you for your replies. Sorry! In hindsight, doing the USB part outside the uC is maybe useful for learning but actually quite boring. I have already messed with other simpler protocols, and while I could read the 200 pages USB specification, there are other things I would fancy learning first.
To get back to the original project, I don't think the actual uC is critical for that task, and I might use the first one I have on hand that satisfies the basic requirements.
What do you think about the sensing method? One big problem I found with commercial joysticks for flight simulation is the lack of resolution around the neutral position, and that the sensors (whatever is used) tend to wear out pretty quickly and become awfully laggy/noisy or just get a dead zone around the middle.
So, what do you think of the three methods I suggested: inductive (the moving joystick would move an iron core inside a coil), capacitive (it would move some metal plates, or I could just use a variable capacitor commercially available) or hall effect (I would put a magnet on the moving part and measure the magnetic field). Someone mentionned potentiometers, but I think they are the cause of wear and play on commercial joysticks.
The general problem is that a joystick for simulation have to be quite sensitive (I don't know if you've already flied a real plane, but on some small ones, just 5mm of movement on a 50cm yoke stick can already have an big impact, and that's just about 0.5°). I have too options: either the sensor is sensitive enough, and everything's fine, or I have to put some mechanical amplifications, and now I have to make an accurate, wear-free and play-free mechanical amplifier, which doesn't seem much easier without taking too much space.
What are your thoughts?