When the tribunal picks cases to be reviewed, it randomly chooses them from a large pool out of fairness to the person being reviewed. Lyte has commented on this in other threads, saying that a typical case can have a pool of 70 games. This would be a pain to view them all, so the system randomly chooses only a few to give a decent snapshot of a player.
What if it only choose the "representative" games, by which I'm assuming the bad ones? Well... that's not really fair to the person being reviewed. What if someone had 60 borderline cases and 5 obviously inflammatory ones? It wouldn't be fair to choose the ones that painted the player in a bad light if he was just extra moody for a couple of games. Therefore, some of the cases may not have anything obviously offensive just based on how the tribunal chooses them. If someone reviewing the case finds two out of five absolutely venomous games and the remaining are "Yeah... that's not ban-worthy but I still wouldn't want to play with this guy," they can extrapolate and assume that there are other bad cases that weren't chosen. It takes a lot of effort to get to a tribunal case, as it has several layers of filtering just to mitigate false positives as much as possible.
Because of the way this works, it's generally safe to assume that if someone even enters the tribunal, they have probably done something ban-worthy. This is where the individual judgments of the reviewers come in: one person may feel that one flagrant case out of five isn't enough when the probability is extrapolated out to 60 games, while another may hit "punish" whenever he or she finds something that breaks the Summoner's code, even if that person being reviewed might have just had an off-day. When enough people vote on a case, these extremes balance themselves out. Overall, the tribunal does have a higher punish rate than pardon rate, but, as I explained above, that's just due to how the system works.
