Ha, so basically tie an actual energy cost to the whole "programmer speed" vs. "execution speed" graph (see).
I think the paper is exclusively talking about execution of the program and not the resources in creating it. Considering programmer time is measured in individuals or small teams and execution is measured in millions of executions/uses, my bet is, for the most part, any cost associated with program creation will quickly get washed out by execution cost. Notice how many Python libraries are created in lower level languages with hooks into Python for exactly that reason.
Yes assuming the code is reused a lot over a long enough time frame, creation cost could effectively count for nothing. I guess we’d need the average lifespan of a piece of software in a particular language. I suspect C programs have a relatively long lifespan so perhaps the original point stands.
reply