I am not completely sure, but from what I could understand, the registered data matches the depth information and relates to the RGB camera.
By taking a look at the /camera/depth_registered/disparity we can see that it is slightly different from /camera/depth/disparity
/camera/depth_registered/disparity seems to be the image got from the rgb camera, showing only the part that is visible by the depth sensor. It is useful when you want to relate the RGB image and the depth points. Imagine you want to 3D scan an object. If you need to know the color of that object, you should use /camera/depth_registered . That is how the /camera/depth_registered/points might be built, as it kind of mix the information from RGB camera and depth sensor.
/camera/depth/disparity contains the raw depth information from the depth sensor, though.