Currently the crosshair is realized by the intersection of the three view planes. This only works in the StdMultiWidget, where all three 2D-views always show the three anatomical (orthogonal) planes.
Using the custom multi widget, it may happen that the three view planes are parallel and therefore no intersection is visible.
The problem is already mentioned in T22876.
Furthermore, the crosshair is based on the GeometryPlanes computed by the different views. If we want to have individual render windows, we might not provide all three views of a loaded image and therefore cannot retrieve the geometry plane of three different views.
The questions remains: What is the purpose of a crosshair / position pointer for a single render window?