Parent is right, it's how it "should" work. Encode for a quality target not a bandwidth target, but with a bandwidth cap (of sorts, generally a virtual buffer average bitrate target to simulate refilling bit buffer at targeted line speeds when depleted by high motion or latency).

Source: Built and operate a commercial transcoding cloud for studios.

EDIT: Added "should". As jfb noted elsewhere in this thread, often poorly encoded content is shoved down aggregators throats by content owners or third parties in the workflow.