I think it doesn't make sense to disallow BLJ and allow this instead. The operating principles may be different as per andymac's words, but the result looks, and in effect is, identical: Mario accelerates to otherwise unreachable speeds allowing him to pass through obstacles he isn't supposed to according to normal game physics.
Innovation by discovery is different from innovation by application. Pretty much the entirety of the last generation of SM runs falls under the latter. The first version of the glitched any% is, on the other hand, an example of innovation by discovery.
Family Feud works well in this category not because of its humor, but because it is a (very) successful attempt to TAS a kind of games that was thought to be impossible to TAS in any sensible way.
I believe it makes little sense to nominate improvements to existing movies in the same categories, since in that case we will be seeing all the same stuff winning over and over.
— Alien Soldier (!);
— Tetris DS;
— Gunstar Super Heroes;
— Castlevania: Symphony of the Night (Maria is the fastest, I guess);
— Strider 2;
— Hagane.
If going purely by the merits of author's input, then it's Family Feud, no real contest here. The second would be Chrono Trigger glitched any% due to how hilariously it was broken, but the difference in entertainment rating shows for itself. As for stylistic choices in non-glitched speed-oriented movies, I'm not aware of anything special this year.
Castlevania: Aria of Sorrow — great pacing, creative soul use, very apparent luck manipulation and movement system abuse. Probably deserves Circle of the Moon's star.
Bionic Commando — the movie looks incredibly polished now, with decent stylistic choices. At the expense of Ninja Five-0, I guess.
Gunstar Super Heroes — blazing fast, flashy, pertaining to ADD victims. Not sure if it should take Gunstar Heroes's star.
Also worth discussing:
Castlevania: Symphony of the Night any% (breaks the game apart in surprising ways);
Donkey Kong Country 102% (just very entertaining);
Chrono Trigger any% (short and confusing, though probably too confusing for beginners).
Possible stars to recycle:
Biker Mice From Mars (breakneck speed gets old quickly);
"100 CDs" Rockman & Forte (pacing problems);
Darkwing Duck (too megaman-ish, considering a lot of starred Megaman TASes);
King's Bounty (it's a great niche movie, but most likely not a perfect recommendation for beginners who search for some "oomph").
Battle of Olympus.
Drang Nach TASVideos!™
The "known improvements" rule should be abolished as a rule because at least 90% of the published movies have had known improvements at the time of submission. It fails as a rule, and this submission is a testament to its failure.
Another point to consider is that controversial rejections should be handled by judges who know at least something about the stuff they reject. "I think I see an improvement here" is such a poor excuse I don't even know what to say about it.
It might also be worth pointing out that, by the time this movie was created, it would have been considered optimal by general standards. For a not-so-well-versed person who doesn't know about recent MM4 developments this is as good as it gets.
What I want to say is that there is no sloppy play, and it makes this case notably different from, for instance, OoT any% submissions that omitted the tricks known at the time due to various reasons that can be combined under an umbrella term "sloppy play".