The major problem with this is ensuring that the cylinders on the platters are all aligned. Current methods of manufacture record the cylinders with their spacing very closely fixed, but their actual position on the platter is not fixed as accurately - this means that what may the the optimum position for reading/writing a cylinder on one surface may not be the optimum position on another.
Cylinders are so closely spaced, that relying on any method of finding them (other than locating them with the head that is doing the reading) simply doesn't work with current drive technology. As an example, the IBM Deskstar 120GXP has a minimum cylinder spacing of 0.45 microns (a size on a similar scale to the size of features on a modern CPU).
Even if you could get it to work, you'd have reliability problems due to uneven heating while operating (and therefore uneven thermal expansion causing the heads to go out of alignment) and manufacturing problems such as the spindle or servo not being mounted straight (A misaligment of a few minutes of arc shouldn't be a problem with current drives, but one of these stacked drives wouldn't work at all - even a deviation of a few seconds of arc would cause grave problems).