The two sizes of tubing I have experimented with both have an inside diameter of .75 inches, with thin-walled and outside diameter of .875, and thick-walled with an OD of 1 inch. Neither is sufficiently rigid without an inner core, the thinner one is more pleasant to hold, the thicker one is more protective. With a hardwood or aluminum core, both are more than rigid enough.
The solid-state gyro+accelerometer is the better sensor, but the mercury tilt switch is much less expensive, easier to interface to, can also operate a sleep-wake circuit independent of the processor. Its effectiveness during staff rotation is surprisingly good; it must be located with the sensitive point of the drop of liquid metal very precisely at the center of gravity. It does not much matter where your hands are, when free to rotate, the object will rotate about its CG. The code discards errant readings by only looking for consistent sequences of on-off timings as in a boolean phase-locked-loop. Thus, having a few missed transitions does not cause a problem, as long as the speed remains approximately consistent, and as soon as you establish a new speed, it will get picked up by the loop.
I have not yet done enough with microphone input to conclusively speak to that.
Of the camera, I did not do the photo/videography, others more capable in those areas than I did, and what I know is that lower shutter speeds are better; orientation of the plane of motion to the focal plane of the camera matters; and if possible, at least approximately sync the rpm to the shutter speed, for the clearest image.