It is unclear how clip should work in 3D. Currently, picking works this way: pick the node normally on its pickRay intersection if there is any intersection with the clip node, regardless of the mutual position of the node and its clip.

Is this the right way? How should clip behave for 3D primitives?

Right now there is also inconsistency with 2D rendering - when a 2D node has a 2D clip which is translated along Z axis, the node is not rendered, but it is picked. After the behavior of the clip is specified, make sure that this inconsistency is removed (one way or other).