This new piece in American Economic Journal: Macroeconomics seems to be channeling some parts of Bryan Caplan's argument:

Despite increases in the college earnings premium to persistently high levels, investment in college education remains low. We can understand this apparent puzzle by considering the risk of attending college and, in particular, the possibility of failing to graduate. Students with a reasonable probability of completing college already enroll, and for those who do not enroll, the low chance of completion blunts the impact of the rising college premium. In the absence of improved college readiness, our quantitative results suggest that continuing long-standing trends in skill-biased technological change can be expected primarily to increase earnings inequality rather than college attainment.

From Kartik Athreya and Janice Eberly.  The implied discontinuity in the de facto talent distribution also echoes some themes from my own Average is Over.