A very belated post . . .
Yes, if the device doesn't "see" the congestion, it's configured QoS, for congestion management, will generally be ineffective.
To deal with later-then-this-device-bandwidth-reduction, as you note, you would normally shape for that bandwidth. Then the device can manage congestion, much as if were the device which had the reduced bandwidth connection connected.
One note, such a shaper would normally manage all the bandwidth to the far end, not just one class. Then a child policy might be used, including LLQ, to manage congestion in detail.
I'm not quite sure I understand your remarks about rose colored glasses, or cheating.