The quality of peer review varies with the journal. For instance, Journal of Computational Physics tends to give out helpful reviews. From JCP, I usually get back a list of helpful suggestions and weak spots in the paper. Even the incorrect suggestions/criticisms tend to reveal ambiguity in my writing more than anything else.

PRL, not so much. In my experience, at least one referee doesn't even finish reading the paper. It's not just me, this happened to a friend of mine last week:

Referee: "The authors do not demonstrate that their equations apply to any physical example."

Response: "See page 3 paragraph 4."