I was always taught, the best learning is achieved at about 70% success rate.
Too high success rate indicates one is not facing any difficulty, and is not learning to improve beyond his current ability.
Too low success rate does not leave sufficient good examples to follow.
(I did not vote because such an option is missing in the poll).
What Amir said!
It corresponds both with what I understand from the research on learning done in both humans and animals, as well as my own experience. NO or very rare failure and you're likely not doing anything you don't already know, there is no pushing of the boundaries, no impetus to reach higher, no signal to point out things you shouldn't do. But beyond a certain amount you're just diligently practicing failure over and over again (and practice makes permanent...) -- which teaches humility and all that, but most of us want to learn something beyond just humility.
No or rare success and there is no signal to point out which things to keep doing -- we learn new things fastest when there is a clear contrast of results where sometimes the results are positive and sometimes not, so that we can constantly compare the two and increase the things that led to success vs. the things that led to failure. Without having both so you can compare (especially on a more automatic reflex level, but also consciously) the results from different courses of action, your brain doesn't really have enough information to know which direction to go next. That's why I would say you can't vote on one or the other, because their relevence comes from the comparison of the two.