Grades can indicate the depth of knowledge, but with the disclaimer that both the arbitrating grader is trusted (in a very special sense), and that the method used to decide grades be both defensible and explicable to others.
To provide a concrete example. A paper on the first book of Hume's Treatise of Human Nature attempts to bring a new argument to bear on the consequences to modern epistemology incurred by Hume's concept of abstract ideas. The paper is written as a final in a graduate level philosophy class, and the instructor is known to be knowledgeable, intelligent, fair, and honest. The paper receives a B+. It is arguable that the mark is generally reflective of the student's understanding and ability at the time the student wrote the paper. (this example is taken from personal experience - I wrote this paper, and I feel that the grade was reflective of my knowledge on the topic)
But without the trustworthy instructor, the grade would be meaningless. The problem, if I were to speculate, is one of numbers - too many students, too few professors. A basic premise of security is that the more parties involved in a pact, the lest reliable the pact is, and if grades are to be a faithfully representation of a students ability, the reliability of the academic pact is paramount.