Now, I've got no idea how Mark plans to scale the scores across three categories and take into account each game's amount of reviews. But this is what makes the most sense to me:

Gooey Castle is the top game in Design, so I'll use it as an example.

Design: 4.890 Theme: 4.890 Originality: 4.918

Add them all together = 14.698

Divide by three (there's three categories) = 4.899

Divide by amount of ratings (in this case 73) = 0.067

Subtract 5 (max rating) and convert to a positive for your scaled average score = 4.932

So yeah, this may or may not be how the top 100 are chosen. I've got no idea, but this is how I would do it.

I just wanted to say that this would reverse the ranks. For example, if they got all 5s (Design: 5 Theme: 5 Originality: 5) then the sum will be 15, after dividing by the amount of ratings (73), it will be 0.205 so the final will 4.795 which is lower than 4.932. This means higher scores will lead to a lower average. Anyway, by the time I wrote this, an overall ranking has already been added which is calculated by averaging the three scores: Overall = (Design+Theme+Originality)/3.

In my mind a mean average is a bit mean :)

Ludum Dare allows voters to specifically designate an overall rating. This allows for something like the following:

Design = 4
Art = 4
Fun = 5
Overall = 5

Which I think is a fairer interpretation of the voter's intent than a mean average of 4.3.

Otherwise voters may artificially inflate ratings for all categories to ensure the game they like does well, which helps no-one.