[remus]

I think you would need to understand in some detail the makeup of all the content on YouTube before you can make that choice.

As a hypothetical example, say yt had one user and 1 billion years of video. It makes a lot more sense for that one user to do up scaling than it does for yt to transcode all that video. The reverse is also true i.e. 1 video, 1 billion users = yt should do a lot of transcoding.

The real solution (in the economic sense of minimising overall cost) is probably a compromise where yt transcode their most popular content, and if you're interested in the long tail of unpopular content then you can invest in upscaling.