Mudge has actually written an excellent followup to his technique to balance type I and II error rates to determine a study specific α. He actually went back and did a meta-analysis to look at how study specific values of alpha can change conclusions – sometimes substantially. The paper is great as well.
For more information, here’s his original on setting a study specific α and a thoughtful piece by Gelman and Stearns where they highlight that the difference between “Significant” and “Not Significant” is not itself statistically significant [pdf]