P(any given 800 ELO player is a troll/AFK) = X

Then the probability that your team has no troll, assuming you are not a troll = (1-X)^4, and the probability that the other team has no troll is (1-X)^5. If you're duo-queued, that changes to (1-X)^3.

So if, say, 20% of low ELO players are trolls, but you are not, your team gets a troll 59% of the time.

Here's the issue: Your opponents get a troll 67% of the time.

Assuming equal P(Victory) in matches with equal number of trolls, and 100% chance of victory with a troll advantage(Which doesn't actually happen mind you), the following results, repeating that 20% trolls figure:

Case 1: You have a troll, your opponents do not. You lose 100% of these under the ridiculously inaccurate conditions earlier set. This comprises .59*.33, so 19% of games. Congratulations, you automatically lose 19% of your games. Sorry to disappoint you, you won't go undefeated.

Case 2: You have trolls, but your opponent does too. These split 50/50 if skill levels are even, and comprise 40% of the games.

Case 3: You don't have a troll, and your opponent doesn't either. 14% of games, split 50/50.

Case 4: Your opponent gets the troll, and you don't. 27% of games, and you win all of them.

So by the end of this, you win 54% of games just by being an average player and not a troll. This is obviously not the actual case, but that has a lot to do with the fact that you probably aren't as good as you think you are. Statistically speaking, 50% of the population has to be below 1200 ELO for it to be a true average.

ELO's a good ranking system; admittedly, Riot could do a better job of forgiving more games where there's an obvious AFK, but still, the problems your alleging solve themselves by immersing yourself in data.