Another idea for steering out tilt on a multi-stage rocket is to have a servo move a mass laterally within the airframe to slightly shift the rocket CG off of the aerodynamic axis. During the burn you would want to use relatively low gains and it would steer the rocket in the direction of the mass shift, and then during coast, when drag is decelerating the rocket, the rocket would move away from the direction of the mass shift. I haven't done the math yet to determine how much mass you'd need to move in order to make a measurable difference, but I'm intrigued by the idea because it would be nearly drag-free and somewhat simpler implementation. One downside is that it could only be used to correct tilt, but would have no effect on roll.