[T]he optimal urban area size is about 750,000 people and the optimal density is about 2,000 jobs per square mile–roughly Baltimore or Philadelphia.
That's from Randal O'Toole.
I did not know that. The Census Bureau has not yet released "2010" census figures for urban areas, but in 2000 there were at least 33 urbanized areas with too many people.
(The paper O'Toole cites does not, in fact, attempt to calculate the population that maximizes productivity but rather the population that maximizes patent intensity per capita. See The Gated City for the extensive evidence that productivity rises with density.)
Ryan Avent has a detailed response to O'Toole here.