After my scene views changes (svn 18628), I can still reproduce this bug.
My feeling is that if the link is restored before the fg/bg volume settings, as each slice viewer tries to restore it's fg/bg, the linking forces an update to the others, so you end up with them all the same.
In vtkMRMLSliceCompositeNode::Copy, the Linked vars are set after the bg/fg but before a few settings, but I think the problem is that during mrml scene restore, the modify events will get triggered as each slice composite node is restored and that will trigger the linked behaviour. Something needs to check the IsRestoring flag on the mrml scene and not link, but I'll have to leave that to Jim.

Any progress on this bug? It makes sceneviews useless.

Looks like the same problem with scene restore.

I set up a 3-over-3 layout, configure each viewer to have different same Bg, different Fg (except first viewer no Fg), all slices were in the axial view mode and linked.

When I open the scene, everything is a mess, as can be seen in the attached screenshot:

1) same Bg/Fg in all viewers
2) orientation changed to Axial/Cor/Sag

Any updates on this bug?

Without this functionality working, there is no point to try to transition our prostate mpMRI review to Slicer4.

This is fixed in r19319.

Added priorities to events listened to by the SliceLinkLogic so that SliceLinkLogic is the LAST to be informed of the EndBatchProcessing, EndImport, ... style of events.

Had to modify the vtkObserverManager and vtkSlicerAbstractLogic to support priorities.