I'm inclined to disagree. Deep abstraction gives us powerful tools for solving less abstract problems, including those that come out of the empirical sciences. Even fields developed with a deliberate eye to avoiding practical applications have sometimes turned out to make significant contributions to the sciences (I understand knot theory, for example, began this way, but has since turned out to have important applications in biochemistry).
You make a strong point, however; the question as to whether we can or cannot improve the efficiency of mathematical research appears to be an open one. I think that perhaps the real issue is that we don't have a correct reductionist account of mathematics, and thus are not able to see clearly what we are doing when we build our theories. If we had a better road-map, I think that at the very least we could tie mathematics down to level 1/level 2 space so that we could have a better idea as to how we can measure the profitability of various possible lines of inquiry.
It looks like, this month, I get to be the one to start the quotes thread.