Technically F90 is always 1.0, the range in which it gets hit just gets narrower.
Here’s a plot of dielectric Fresnel reflectivity for various IORs (note the nonlinear X axis):
They all go to 1.0 eventually, but for low IORs you need a VERY shallow angle.
And since I already have Gnuplot set up, here’s the same graph using the Schlick Fresnel approximation, which is very widely used, especially in realtime engines:
Note how it works well for “normal” IORs, and gets the same F0, but the behavior at high angles is completely wrong at low IORs (as you might e.g. get for nested dielectrics, or when putting the IOR adjust near zero of course). Therefore, a renderer that parametrizes purely in terms of F0 and F90 won’t get the smooth transition to diffuse.
Hm, it could be done, but the main difference between the main specular layer and the coat already is that the main specular layer gets tinted by the coat. Making the coat also tintable seems somewhat redundant. Of course there will be cases where it is useful, but that could be said for many options.