I do not believe this would work against a determined (and very smart) attacker. Although video files test as being completely random, I believe it is quite likely that there are patterns within them. See this comparison of overheads for containers
. In a ~993MB MP4 file, the MP4 container is about 2.4MB. Not much, but it seems that MP4 structures would appear at predictable areas in the file. (MKV is 1.5-1.8MB, btw)
Additionally, it is easy to figure out if One Of These Things Just Doesn't Belong with a video file. Just demux the video file. (There are many, many tools to do this. FFmpeg can.) If the video/audio tracks add up to less than the actual filesize (minus approximate overhead), then there is something odd about that file. Or, obviously, if one of the tracks is of unknown type and you can't decode it.
But I am assuming a "We only have twenty four hours to disarm the bomb and the code is in this computer!!
" situation. Against a less determined, or more cursory, inspection (ACTA's border searches, for example) this steganography would work great.