It's difficult to find average expectations for performance in a population, but given enough interactions it will be found. The problem is that when any kind of learning is needed within a group without any expectation of past or future social interactions, benefit depends on a prediction of the amount of effort to reach the goal but obtaining the information to accurately make that prediction takes its own effort, which may turn out to be a lost investment.
Someone may be presented with two possible cases: 'competence', or 'lack of competence'. In the second case due exactly to the factors leading to lack of competence, advice for improving mistakes may not be instantly accepted. At the same time, the system must be designed on the assumption that improvement will happen.
. . .
Meuh, fine, this entry is about WoW now—
It's important to separate players making honest mistakes from lack of experience from those who blame their mistakes on others and are unwilling to improve.
If the playerbase cannot assist those of the first category, Blizzard will nerf the game, making all of the current encounters much easier exactly like you imply you dislike. The second category is not important, but the only way for a group to distinguish between them is iteration; it can't be done instantly the way groups would like.
For accepting risk to become a popular decision, one of the possible results of taking greater risk must actually be seen as rewarding.