This article seems to completely ignore the biggest area where inline CSS is problematic: dynamic, state-dependent styles. That is, things like pseudoclasses (hover states and the likes) and media queries.

You can do these things in JavaScript: match your media queries in JavaScript and change the inline styles, and reimplement :hover, :focus and the likes. But I expect (without supporting measurements) that doing this for everything will be fairly bad for performance, and it will generally be less responsive (in the performance sense of the word rather than the resolution-independence sense of the word), e.g. your state-dependent styles may apply a frame later than they should. And I don’t like to depend on JavaScript without good cause.

This is where utility CSS frameworks strike the balance: by using regular class names for the effects of inline CSS, they can apply media queries and psuedoclasses. It’s an interesting approach that I haven’t used yet but I’ve steadily been coming around to the notion of for some sorts of projects.

(There are also pseudoelements; those can’t be implemented with inline CSS at all.)