The average of nothing and something is... something, I guess.
In other words, if you averaged a part of the JSC where there was no picture information (all black), with the part of the 2004 DVD where there *is* picture information, then all you'd get back is the 2004 DVD picture. And I'd be inclined to think you'd then see the edge pretty clearly.
Even lining up the 2004 DVD and the JSC would take longer than the day or so X0-1138 purports to have been working on this.
On a vaguely-related topic, one experiment I've never tried is to use ColourLike() to give an HD target the same colour palette as the JSC.