Haven’t had enough coffee to think about this rigorously.
My intuition says that as long as you could get to the desired 3D shape from revolving the 2D shape around an axis, essentially integrating the area into a volume, the results will be valid or equivalent.
I don’t think that’s the entire story, there are probably other ways to simplify 3D shapes. And yes, onions will have non constant variations (or ones that don’t cancel out to 0) along the sweep which is what actually invalidates the real world application.
If you model the (half) onion as a stack of these slices, it’s clear that the radius of each slice varies over the height of the onion; so the points below the onion found by this method towards which you need cut will form a curve, not a straight line. That is hard to accomplish with a straight knife that makes planar cuts.