Answer:
Each one is larger, cheaper and slower than the next. L1 cache tends to be direct mapped and L2 tends to have higher associativity. I don't believe that any current processors use L3 cache yet, but it would likely be fully associative or almost fully associative.