There is no method that is perfect. It often boils down to the equipment you have on hand.
With EDXRF there can be particle size problems. We handled particle size effects by embedding the titanium dioxide in a suitable matrix-usually wax or plastic.
And the calibration curves can be interesting. Most people expect a Beers-Lambert type curve. As you alluded to, the shape of the curves can often be different from a typical Beers Lambert curve.
To me, this is offset by the advantage of not having to handle digestion in strong acids. I don't have ICP, only AA. The special nebulizers and reagents needed for AA aren't worth it to me.
As usual, each chemist has to weigh the advantages and disadvantages (usually instrument cost or instruments on hand, analysis time, and reagent cost) to make an informed decision.