The magnification will indeed depend on your reversing hardware. The
more distance you put between the lens and the body, the higher the
magnification. The exact formula is:
magnification = lens_to_sensor_distance / focal_length - 1
The problem is that the distance from the lens to the sensor has to be
measured from the relevant
principal plane of the
lens, i.e. the object side principal plane, which becomes image-side
once the lens is reversed. Then, to compute the magnification, you need
to know the position of this plane inside the lens. Alas, I have never
seen this information published for current lenses. You may be able to
compute this position yourself... provided you have the complete formula
of the lens! Finding the necessary data may be harder that finding
someone you could borrow the lens from to do an actual test.
As for the question of what is practical... assuming you can achieve
unlimited magnification (bellows, etc...) you will likely be limited by
the resolution getting bad at too high magnifications. You can expect the
MTF50 of the lens to be roughly divided by the magnification. Then the
maximum practical magnification will be:
max_magnification = lens_resolution / required_resolution