YouTube uses some kind of bursting heuristic that is marginal in certain scenarios.
What it does is it loads the first chunk of a request to start playing with high bandwidth, but it then throttles the connection to approximately the bitrate of the video stream. The idea is that YouTube reduces the costs (bandwidth, CPU, routing etc.) of transferring bits that are never actually consumed, because the user stopped watching the video before it finished playing.
The quickest way to get a YouTube video is to download it with a download manager that sends multiple requests. Those requests ought to trigger the initial burstiness, and if you have enough of them either the initial burst will be sufficient to cover the whole file or the total bandwidth (even though throttled) across all simultaneous downloads will reach your own download bandwidth.
What it does is it loads the first chunk of a request to start playing with high bandwidth, but it then throttles the connection to approximately the bitrate of the video stream. The idea is that YouTube reduces the costs (bandwidth, CPU, routing etc.) of transferring bits that are never actually consumed, because the user stopped watching the video before it finished playing.
The quickest way to get a YouTube video is to download it with a download manager that sends multiple requests. Those requests ought to trigger the initial burstiness, and if you have enough of them either the initial burst will be sufficient to cover the whole file or the total bandwidth (even though throttled) across all simultaneous downloads will reach your own download bandwidth.