If you have a cubic baseline with '-polort 3', then any cubic trend
in the motion parameters (or your regressors of interest, for that
matter) is irrelevant.
To put it more generally, adding a multiple of one regressor to
another will not affect the second regressor's beta weight, it will
affect the weight of the one added (constant polort in this case).
Of course, that generalizes to a list of regressors, such as the
four for a cubic polort baseline.
Again, this assumes the regressors are linearly independent.
----
Modelling motion parameters separately per run is different than
modelling them across all runs at once. In fact, afni_proc.py has
an option to do that (-volreg_regress_per_run).
Modelled per run allows the beta weights to change per run. When
modelled across runs, the single beta will be closer to the average
of the per-run betas. And of course, the single regressor will
account for less variance, leaving more "residual motion" in the
data, say.
- rick