Well the main problem with the use of root motion is there is not a root motion movement component like there is as to in-place and the simple fact that using root motion or not the “character” model is “always” in a constant state of falling. The benefit of course is an off the shelf ready mix solution that solves the problems and needs for a constant as to the needs and demands for controlled input with in a networked component that takes care of the replication and error correction requirements with out having to do it from scratch with in an all RM design.
The result though, and as you mentioned, is the character model becomes un-grounded that would turn the result into a user imputed Pixar movie so it’s still the same old same old of having to author a hack work around to get proper movement dynamics.
I went on a bit as once the character is “in air” then control is returned to the capsule so any kind of in-air control would have to be applied to the capsule as in “Is Flying”.
So no I really don’t see anyway of controlling world space movement using RM once the player leaves the ground with out having to rebuild the frame work from scratch.