Ahhhhahahah.... you are assuming there are players with 100 or more games who aren't terrible players.

You are assuming there aren't players with less than 100 games who can't rape 95% of the DG population.

You are assuming there are enough players at once to get even 1 game going on the different servers.

I am afraid that i also mentioned about %

You can't always trust % or games played. I've seen players with 100+ games with over 50% who still sucked.

Rating = w(W * 10)^2 + 2(G) 

G = number of games

w = total wins

W = Win percentage (as a decimal)

Base game matching by rating, don't allow doubles, don't give a rating until at least 10 games have been played.  A more complex formula would be better, but IMO the number of games someone has played shows skill better than win %.  

100 games, 40%

40(0.4 * 10)^2 + 5(100)

=40(16) + 500

=1140

50 games, 60%

25(0.6 * 10)^2 + 5(50)

=900 + 250

=1150

I think this system would be fair in evaluating win percentages vs. total games played, while encouraging new players to play more games, as you rating should increase the more games you play. At higher win percentages, your rating would grow exponentially as well.  Players could be grouped by rating and ensure each team has a balance of generals and assassins.  (Players should be prompted for an optional second and third demigod, and be able to choose their map preference and game size preference.) Of course, until more players join Demigod, some of these preferences could not be adhered to strictly without limiting games.  

but shut up now

now those premade noobs are all gone

But since the stats are reset... Not anymore really

people(noobs) often just don't play in premade room unless they are premade too