You are perfectly right, each time there is a sharp corner, the shell tool has to “guess” what to do at this corners because it is not possible to keep a constant skin width at such a summit.
This is what surprised my yesterday because I had never thought about that before, and it is not that trivial in the end.
When it is straight and simple geometry, mostly made out of lines and circles, the algorithm handle it the way we expect it, because, regardless of the maths, we need to be able to shell usual bodies 
.
When it is more complicated, using spline, the tool does what it can, and when it fails, we can help it by adding rounded edge for instance at the sharp summit of your shape, or by changing the shape of the spline near the summit.
Video below with rounded summit
(the last two control points are on a line perpendicular to the revolve axis)
Negative shell = outward : the tool keeps the skin’s width constant by creating the large radius summit
Positive shell = inward : the tool has to compromise with the request at the summit; the rounded edge there helps a lot (from my experience).
Again not that trivial when you deep dive into it, and I’m sure that we’ve just scrapped the surface.
Pec