The argument is that in the real-world setting in which people actually use interfaces, they are distracted with multiple stimuli and trying to perform multiple tasks at the same time. They say, “they’re devoting, at best, 10% of their attention towards your app”.
And “drunk people are a pretty accurate mimic of distracted, indifferent people.” Therefore, why not take your user testing into the bar for a cheaper and possibly more representative test?
Clever thought. For the price of a $5 beer, it may not hurt to give it a try.