I'm not sure they should sample the three best games. Instead, maybe just factor in all games played? The problem with that though is that some people would play just the minimum games and hope to get a small, but really potent selection of games to average from. It'd be the race to 50 but every month. Players would secure a good rating and then not play to "secure" a good score.
That's what's nice about the individual performance based ratings, which are independent on whether you are on the winning team. However, a good team lends itself to better individual ratings.
But what about role-players such as: sniper, close quarters, bait, mid-range, etc...where players may not all have the most kills and most of the time the sniper/ranged players stay back safe and rack up assisted kills from sacrifice players?
HERE'S A SCENARIO:
Assists are defined as dealing at least 40% of the player's damage before they're killed. Well, what if you have a great team who coordinates and team shoots the crap out of some guy with DMRs? Let's say each player shoots the enemy just once, simultaneously. In a 5-shot situation, they'd each be dealing 20% of the damage. The final 20% will go to someone, but is anyone awarded an assist? The first four shots landed will pop shields and the final head shot/body shot for the kill will be a toss-up among 4 players all shooting the same target. In terms of rating, the lucky guy who lands the killing blow in an otherwise truly assisted situation will be helping his rating, while the TEAM will gain nothing. Now extrapolate that scenario over the course of a game and season. Of course, not every kill will have 4 players shooting the same target, but in concept it's the same as if 2 or 3 guys are shooting the same target.
It is possible for 2 people to have assists for the same kill if they have 40% each and the last goes to someone else.
Bungie, if it's not a factor yet it should be: landed shots(accuracy) should count in some small amount toward rating.