From a new paper by Derek Pyne in Economic Letters (December 2012).
Empirical studies have found that increasing the probability of punishment has a greater effect on crime than the severity of punishment. This note explains this as the result of criminals having imperfect information on their criminal ability. As they commit crimes, they update their estimates of their ability, based on their success rate. Increased penalties deter crime in the period they are applied but offer criminals no information on their criminal ability. Crime is also deterred during a period of increased enforcement. In addition, increased enforcement leads some criminals to decrease their estimates of their ability, leading to reduced recidivism.Alas, I did not find the paper online. There is some more here.
This finding is "close" to the "broken window theory" of crime, which was a guide to the policies that reduced crime in NYC.