...if you took a sphere made of iron, divided it into regular pyramids, with the apexes at the center of the sphere and the bases on the surface, magnetised each pyramid so that each apex was the N pole and each base was the S pole (or visa-versa), and then super-glued it all back together again?

Obviously, the difference in the amount of magnetic material would be uneven along the length of the magnets so we could refine the experiment by making each pyramid out of a matrix where the domain density over length, but not volume, remained constant - imagine pouring a suitable resin/domain mixture in to an inverted pyramid shaped mold where the flow rate of the resin increases with time but the amount of domains mixed in with it does not.

The next step would be to use electro-magnets instead of permanent magnets.

Just an idle curiosity question:)