what's the difficulty in splicing in some additional frames at upload or request-time? From what I vaguely recall of the format it's essentially a stream of stream of independent frames with some timing info, although there might be something about colour palette selection? I'm sure there could be something that makes it especially horrible, but curious what it is :-)
Isn't video advertising mostly handled at a higher layer by sticking some segments into the playlist/container rather than actually encoding it into a single video unit anyway?