Interesting link, thanks for sharing. I think there's no such a thing like subtractive color space - all CMYK profiles have LUTs witch colorimetric characterizations of various color combinations, so they use a brute-force method to predict what result we'll get when we mix paints.
It's easy to create a subtractive color model that is essentially the compliment of additive models. It's something that's well known and has been around for a long time. Ducos du Hauron was granted a French patent in 1869 for describing subtractive color for use in photography and screen printing.
The problem, and the reason we need to jump through so many hoops to get good color from our printers, is one of manufacturing, technology, and the vagaries of the real world. Substrates scatter light in unpredictable ways; we can't make perfectly absorbing block dyes without unwanted absorptions; etc. So we need to go through a lot of trouble for the extra bit accuracy. But it doesn't invalidate the theory any more than inaccuracy from friction invalidate Newton's. There's a good discussion on the subtractive models and their problems in chapter 7 of Henry Kang's 'Computational Color Technology.'
I think article is pretty murky, but I shouldn't be so critical of the developers. Modeling what happens when you mix paint IS pretty challenging and isn't easily described using either simple additive or subtractive models. Having said that, Kubelka–Munk isn't a rediscovered breakthrough from the past—it's described an Wyszecki & Stiles and commonly used in industrial color matching.