Because of the way your first and third coils overlap you end up doubling the gap and weakening the field.
You'll be ahead if you do a single layer of coils, with the width of each coil's radial runs each being half the width of the magnet face with the gap between magnets also that width. This "paves" 2/3 of the magnet-swept area and lets each coil be the thickness of the entire gap (minus the physical clearance), rather than half of it, thus collecting twice the power. You get about 33% more power with this non-overlapping 3-phase setup than the half-thickness stacked kind.
(You can get better paving with the overlapped kind by filling in the gap in the middle of one coil with winds from another. But it involves coils that are z-shaped when viewed from the edge. Those are really hard to construct.)
Set the gap between the magnet poles and the opposite rotor to about the thickness of the magnetic material for the best tradeoff of strength vs. room-for-copper.
(To get the geometry dead on you'd need trapezoidial magnets. But your layout - with the gap at the outer radius of the magnets being about half the width of a magnet face - is a good approximation. Just treat the little triangle of the magnet that encroaches into the gap as if it wasn't there.)
With a non-magnet-carrying second rotor the field will spread out more than is optimum. You can prevent this by mounting steel pole-pieces opposite the magnets, with a thickness at least that of the magnets - "faking" the second set of magnets. Then the field will go pretty much as straight across as it does with magnets on both sides (though using steel instead of more magnets means you don't get the extra flux of the non-existent second set of magnets.) Of course you still set the gap to about the thickness of the magnet material, which means it's still half the gap you'd use if you had magnets on both sides.
