Not got hand on the IT manuals for the FMC, but I believe from crew experience it has to do with no hard speed/altitude being defined.
Let me explain:
The FMC builds its descent profile backwards, from RW to a calculated T/D.
The final approach segment starts with a point which is coded as 165/1500 (for instance) where 155 is gear down/flap 15 speed for forecast landing weight.
If however such point has its restriction removed, essentially the aircraft does not have a final approach segment defined, even if the altitude is changed to 1500A to accommodate CDA this phenomena can happen; cold weather corrections for the same point like 1590 have no effect on this though.
When the point IS available, VNAV PTH will reach the point from a intermediate approach segment (flap 5) prior to final approach segment. This segment is defined as the waypoint after which the G/P angle is intercepted towards the final approach segment. Hence a 10nm final approach equivalent point in FMC such as (175/2000A) will be used by VNAV for such point. Equally such point could be defining a base turn, however.
If VNAV PTH is unable to define these points, is cannot gradually reduce the speed to UP from defined descent speeds flown, and it then sets the bug to achieve a ~10nm final at flap 5 speed. As the vertical path is not completed thus cannot correctly be created, the information for it is lost causing this anomaly.
Basically VNAV is operating in a degraded mode, trying to achieve what it is designed to do, but without all the information required, to make a perfect CDA.