As I said before, this will remove high frequency components, but you will still see plenty, plenty of low frequency noise after averaging. The only way to get rid of that is to average over an even larger amount of samples, but this will greatly affect the speed at which the system can respond to change, resulting in sluggishness of the controls. And even then, very low frequency noise will still survive unaffected. It's an uphill battle, and it's not serving the purpose. So although averaging is a valid technique when it comes to suppressing high frequency noise in digital signals, it's not what you want when you are dealing with human control data. People just don't like readings to flicker between 39, 40, 39, 40 etc. as long as they're not touching the knob. They will gladly be cheated, as long as the flicker stops!
The "dead band" method is a good way to really get rid of noise on control data completely without introducing any sluggishness whatsoever. I will IMMEDIATELY agree if you say that this is only a form of "cheating", but it is one which suits the purpose perfectly, and therefore it's a perfectly valid engineering solution.



Deadzone constricts the range a little by introducing a nonlinear response. "loose resolution" sounds a little drastic to me. You only loose 2 or 4 steps on the entire range.
Remember that averaging also reduces resolution, but in the time domain.