On unit costs.

The more I think on that 'unit cost' study, the more I think the whole premise is stupid, and you've got the right angle with the 'marginal cost' approach. How much the last 10 units cost is irrelevant, compared to what the next 10 units are going to cost. Adding the first 10 lines to I-15 was cheap, because it was (mostly) rural land. Adding the next ten lanes is going to be very expensive, because it's either going to require: a) condemning hundreds of acres of urbanized land, or; b) decking the freeway. So how cheap driving was/has been is irrelevant.

The kind of business thinking this entails isn't relevant for infrastructure investment. Infrastructure is much more about making major capital investments, ie: investing in a new factory. When you do it, you do so not just for today's needs, but for room to grow. It might be more economical to go with the cheaper alternative now, but in the long-term, the cheaper alternative is inferior.  

Take this example:

 

Marginal cost looks great for cars, but the cost per passenger mile is exponential for VMT, while it's linear for trains. So trains don't look great today, but...we aren't making infrastructure improvements for today. We are making them for the future. And we use 2050 as our planning horizon, and we use that as a cost midway point between 'now' and 'forever' to make economic evaluations.