Land et al recently evaluated the effects of the death penalty on a monthly basis in Texas. They correlated the month-to-month fluctuations of executions with month-to-month fluctuations in homicides between 1994 and 2005. Their two preferred models (i.e., ones that minimizes cross-correlation between the transfer function and noise) have deterrent effects of 2.5 homicides and 0.5 homicides per execution.
Looking at the 2.5 homicide reduction model's function (p 1031 if you have access), I don't buy the displacement effect which postulates an increase of homicides 2 months after an execution. If we are going to do analysis at the margin, we have to consider what the effects of an execution will be on the likelihood that one particular potential criminal will murder.
Psychologically, the postulation of the model is that a would-be murderer would (consciously or unconsciously) be less likely to commit murder for one month after hearing of a recent execution, but then forget about it in the second month and actually increase in the likelihood of murdering. The authors call this "displacement." Then in the third month the probability is back to around baseline and finally in the fourth month there is once again deterrance, although not as strong as in the first month. Sorry, but this doesn't make any sense.
Nevertheless, if you favor the model of state governments as vehicles for policy experiments, then perhaps you have to laud Texas's iconoclastic tendencies, morbid as they may be.