I'm not from the ESRI frontier, but from the 3D OSM side, so I try to answer:
The transformation process for 3D building outlines is a bit complex and depends sometimes on picking the right values: http://wiki.openstreetmap.org/wiki/Simple_3D_Buildings
Thus I recommend to use an existing 3D exporter for OSM, like OSM2World (Java). Sadly there is currently no highlevel export like CityGML that can be used for importing to CityEngine. So you might combine both just in the 3D modeller itself http://wiki.openstreetmap.org/wiki/3D_animation
Please keep in mind, that OSM has a copyleft license that might affect the mix.