Problem was that insert-order (enforced by the optimizer) did not handle
the case where the bitmap changed to a new one.

Fixed by remembering the last bitmap page used and to force usage of this when inserting new rows