There are players that try to keep the teams even because having a whole bunch of stacked games is not fun, and then there are players that stack games on purpose to influence the stats. If a player does either of these, the stats are screwed.
A large enough sampling should substantially decrease the significance of a single player's skill or (temporary—as in the case of "not trying" against opponents who are not as good) lack thereof.
That would be true if people did not make an effort (consciously or unconsciously) to change the balance of the teams (ie. if auto-select was forced). It also would be true if this happened only in a small proportion of games. However, nearly every game is effected by this to a degree (newbies usually pick humans), and I think most of the names I recognize (Cadynum, Garoth, etc) fall into one of the categories I mentioned. If even half (a low figure IMO) of the US server games are effected in this way then we have a significant problem with the stats.
Poor building can also potentially screw up stats, especially since its easier to screw up building on humans than building on aliens (a potentially source of skew away from humans).
If building for humans requires substantially more skill than for aliens, that impacts the balance of the game and should be properly reflected in the stats.
It should most certainly not be reflected in the minds of the developers, or at least not as much as the stats are now. And if it is not to effect the devs' decisions, then it should not be reflected in the stats (if we are going to even use the stats). People adapt to changes, but don't do so instantly. I believe that was an argument the devs used to keep the new turrets around, instead of reverting it immediately after people started to majorly fail at a building. In fact, I think this was taken into account for a lot of the changes. This means that, over time, the balance will shift. I want 1.2 to remain 'balanced' in three years. We should assume that people don't stay newbie builders forever, and that good players are able to /callvote denybuild (or designate when lakitu's qvm is ported to 1.2) so that newbie builders don't cause a 7 minute game-over every fourth game.