An analysis which takes into account the 3 dimensional aspects of the rollers seems to indicate that the correct equation for the profile is parabolic rather than elliptical.
The analysis is given in the presentation titled "equations for bump-free mecanum roller profile" located here:
http://www.chiefdelphi.com/media/papers/2390
[edit] A one-page PDF titled "roller profile parabola vs ellipse comparison" at the same link compares how well a least-squares fit parabola vs ellipse matches a set of XY roller profile data.
Constructive criticism and questions welcomed.
~