The old documentation implies that the call is only valid for the next parallel region and must be called again if addtional regions should be affected as well.