Certain texts refer to DCM and CCM as "complete energy transfer mode" and "incomplete energy transfer mode" respectively, and this is one case where such a description is useful. In DCM, energy is taken from the input and stored in the inductor during the switch ON time. During the OFF time, all of that energy is delivered to the output. Thus, when averaging over a switching period, there's no delay in energy transfer. In CCM, only a fraction of the inductor's stored energy is delivered to the load each switching period. It's this partial energy transfer that permits the RHPZ to manifest in boost and buckboost converters.
Now, if we want to look at things in more detail, we can observe that in DCM, increasing the duty cycle does delay the transfer of energy to the output (since the ON time has been increased), though by only a fraction of a switching period. The standard formulation of state space averaging does not account for when something happens within a switching period. I believe the paper cited in TI's app note is trying to "fix" this. However this sort of behavior is much more complicated than a RHPZ.
edit: and I think this is exactly what is shown in asdf44's waveforms above. It quacks like a duck, but I don't think it's a duck.