If a model produces an accurate formula it does not mean that it is the best model around. And from the linked wiki article I am not sure if I would say the value for alpha/g is predicted by QED. They say that the best models contain these steps: "...by fitting an experimental measurement to a theoretical expression (including higher-order radiative corrections) that includes α as a parameter." and " ...includes QED diagrams with up to four loops. Combining this with the experimental measurement of g yields..". For me this is exactly what Mills critizises about QED in the first chapter of his book. This definitly sounds like curve fitting terms with a half baked theoretical justification to the measurements (after more than 60 years of QED "half baked" is not the correct expression - overcooked is better.).
Mills also uses these pictures to underline that his theory is correct. So I am not sure if these pictures contradict Mills model. I challenge you that you proof that these pictures make a case for QM. The theoretical orbitals in c) and d) are just iso-areas that contain the electron with 90% probability. Combining the probabilitydistribution of the electron (in QM terms) with the measurement uncertainty of the "camera" I doubt that you would get these (more or less) sharply seperated atoms. Mills model has a sharply defined orbit and the "round" (not spherical) elements of the molecule stem from the modulation of the current density function of the "orbitosphere". I think you have to study these measurements in much greater detail before you can come to your (or my) conclusion.