It sounds like the start of a bad joke, but in this case, the researchers were Michael Lyvers and Juliette Tobias-Webb. They recently published a paper in Addictive Behaviors in which they escaped the laboratory and conducted an experiment in a local bar using PEBL's version of the Wisconsin Card Sort.
I refer to the test as "Berg's Card Sorting Test", after Esta Berg, who originally developed the test with primates in the late 1940s.
The test is thought to involve a number of higher-order reasoning skills, and so it would make sense that ability to perform the test would be impacted by how much someone has had to drink. So the researchers recruited participants, gave them a breathalyzer to measure their Blood Alcohol Level, then administered a number of tests on them.
A few methodological details: the BCST task was given twice: once with no instructions, and once with simple instructions that helps people understand the 'trick' of the test. Furthermore, you can measure both perseverative errors (responding incorrectly according to the previous rule) and non-perseverative errors (responding incorrectly for some other reason). The paper shows a negative correlation with performance (the higher your BAC, the worse you performed), which stemmed from increased perseverative errors, but not non-perseverative errors, during the second round of testing.
What does this mean? When you aren't drunk, you can incorporate information (either from the first round of testing or from verbal instruction) that can prevent errors in general, and specifically allow exploratory behavior in the face of negative feedback. The higher the BAC, the less able subjects were to learn from this information and avoid continuing their response pattern too long.