Cost-Benefit Analysis: What is the correct discount rate? Growth rate?

I was working on a paper with Chris Redfearn last week, and he raised one of the most interesting puzzles in economics: how to we think about environmental/infrastructure issues in the presence of discounting? With a real discount rate of, say, 3 percent, a dollar of benefit to our great-great-grandchildren, who might be born 100 years from now, is worth only 5 cents to us in present value terms, assuming that a benefit to our progeny is equal to a benefit to ourselves.

This creates all kinds of policy problems. The example Chris gave is that on an ex-ante basis, removing lead from gasoline did not pass the cost-benefit test. Yet I think almost all of us are grateful for unleaded gasoline. Similarly, I find myself puzzled every day when I ride the Washington Metro that the system does not pass the cost-benefit test. Washington is becoming an increasingly difficult place to live as it is; it would be hard to imagine what traffic would be like in the absence of Metro (where the cars are jammed to capacity during rush hours).

Maybe the problem is that we get potential growth wrong. If people's earning potential increases 2 percent per year, then maybe the way to do cost-benefit properly is to use a Gordon Growth setup with an r-g term: we discount net of growth. Such a change would increase the present value of the dollar to our great-great grandchild from 5 cents to 37 cents.