OK having seen the source code for the performance test and running it myself, it does seem fair for what it's meant to do, which is draw a lot of semi-transparent blue squares all over the stage.
So if you're writing a game with a lot of translucent sprites which need to be made translucent on the fly and not have the alpha already adjusted in a source image, this is definitely a considerable performance benchmark.
Mostly relevant to games which use particle systems, water effects and such other graphical eye candy, I think - which is definitely something fun for artists to play with.
If that was the case, then they chose the worst possible way to do it in GM. Meaning the test wouldn't be fair at all.
In Ashley's words:
The test isn't simulating some particle effect, it's trying to simulate a really big game.
But it didn't do this, either. all it proved is that an empty Construct 'object/sprite' is lighter than an empty GM object. These are similar, but NOT equivalent things. That's a pretty moot point when it comes to actual game logic, and collisions, etc etc etc. The apparent 'heavy' ness of the GM object compared to the C2 sprite is more than made up with then.
So it does prove something, but in the end, it's apples&oranges. Others call it a meangingless test, I wouldn't necessarily go that far, but it certainly the impression that C2 is way faster than it's competitors is completely inaccurate.
The Scirra team are out to prove their the fastest. The only way they can really conclusively do that is by creating some amazing tech demo's that blow their opposition away. They have some nice stuff, but so far, it's only on par with GM-html5 'at best'.
*doesn't know anything about stencyl, gamesalad, etc, so no comment there*