Ok, I've narrowed down the behavior. If I select pin 22 in unit A and move it, then pin 11 in unit B will move. Likewise moving pin 6 in unit A, will also move pin 21 in unit B. The coupling also works in reverse. But no other pins are affected - at least without making any other changes.
Opening the *.lib file, it looks like the pins share the same co-ordinates, only they are in different units. Eg. the unit field is the 3rd from the right - eg. 1 or 2 for unit A or unit B, but otherwise the coords and representation are the same.
X IOR_38 22 650 200 150 L 50 50 1 1 B
X JJJ 11 650 200 150 L 50 50 2 1 B
X IOL_4B 6 650 -400 150 L 50 50 1 1 B
X GND 21 650 -400 150 L 50 50 2 1 W
So it looks like the hit-testing/focus criteria is not adequately discriminating - and just uses the extents without regard to which unit is in the current view.
Perhaps, if nobody else recognizes or has seen this behavior before, then its just an edge-case bug, and perhaps even limited to the version I am running.