Similar to the case described here for the mass origin, I strongly suspect that gazebo does not handle the origin tag correctly when trying to shade COLLADA models. Every time I use a COLLADA model in a .world file or urdf and set the origin to be different from zero (be it xyz or rpy or both), the shading does not work as expected. I suspect that gazebo shades these models as if they would still be at 0/0/0 0/0/0, even though they have a non-zero transformation. An example can be found in the hectorquadrotorgazebo stack and spawned into a running gazebo world using

 roslaunch hector_quadrotor_gazebo spawn_quadrotor_with_laser.launch

The Hokuyo UTM-30LX model mounted to the quadrotor exhibits some very strange shading behavior, for example, when moving a point light up and down, the side of the LIDAR changes color. The model is available here. Normals appear to be correct in Meshlab, Blender and other viewers. It is also shaded correctly in rviz (even if it appears too bright due some material oddness).