In addition to the already mentioned performance issue with client-side MathJax, having native MathML makes it conceptually far simpler to do more complex things with equations, both for the end user and for the developer.
For instance, as a user, if you want to scale the equations by some amount or use a different maths font, it's a couple of lines of CSS, using exactly the same method you'd use to make any other changes to the appearance of a web-page. (Yes, you can easily do the former with MathJax, but I don't think the latter is possible user-side).
As a developer, if you'd want to interactively highlight parts of an equation, for educational purposes, it'd be trivial with MathML, but rather hard to do nicely with MathJax (statically coloured elements are possible with MathJax, with the "color.js" extension, but not dynamically coloured ones — and no, swapping out the entire equation to make colour changes is neither nice nor scalable). Alternatively, if you want to embed equations in a diagram or a graph, it's pretty easy with MathML[0][1], but would be difficult otherwise.
Obviously, all of the above is in principle possible with JavaScript implementations, but it's far harder. You might argue that this extra effort is worth the smaller attack surface. IMO, given the importance of maths and science, it isn't.
Also, why do we, say, have the CSS flexbox layout? After all, we could have used javascript to arrange elements into an appropriate table or even just set the x and y positions of all elements...
[0] http://fred-wang.github.io/MathUI2014/demos/2-mathml-in-svg....
[1] http://fred-wang.github.io/MathUI2014/demos/6-mathml-in-webg...
