Illustrator let's you keep things in vector in the 3D filter.
After applying the filter, you expand - which in corel language is "break apart". You have all the highlights as vectors etc.
You can then apply what is referred to above as "gradient meshes" to the individual or groups of highlights.
This would be a truly scalable vector image (maybe only within AI though). I think when you create the PDF from AI, it will convert all those vector data for the gradient mesh etc to a raster image (I'm 99% sure about this).
In practice, I find it easier to just create the 3D object in AI, without any bevelling (AI bevelling sux and keeps self intersecting paths), apply some extrusion, and perspective. Then I bring that into PS, and start applying effects as above (blending options/effects).
Work at a size bigger than what you need just in case but usually at 150dpi at the width of your printer is more than enough for most things you'll make for the client?