my 5 cents:
Regarding what you write above, fact is that in most papers in the literature the center of mass is used as a reference point when calculating the dipole moment of a protein (or any large molecule with a net charge). The center of geometry might be a better idea as you suggest in your example, but I have not seen this being applied anywhere in the literature. Also, there is at least one more alternative, the so-called center-of-diffusion, see:
not sure how much difference this makes, but the authors claim that this should be used.
Anyway, my impression is that the gmx dipoles tool could perhaps do with some revision - I just posted this: Gmx dipoles and index groups vs molecules
which suggests that there might be undocumented limitations of this tool.
michael