Dan Hurley wrote a lengthy New York Times piece covering the origins of the n-back training exercise, which purportedly improves fluid intelligence in those who practice it daily:

The study, by a Swedish neuroscientist named Torkel Klingberg, involved just 14 children, all with A.D.H.D. Half participated in computerized tasks designed to strengthen their working memory, while the other half played less challenging computer games. After just five weeks, Klingberg found that those who played the working-memory games fidgeted less and moved about less. More remarkable, they also scored higher on one of the single best measures of fluid intelligence, the Raven’s Progressive Matrices. Improvement in working memory, in other words, transferred to improvement on a task the children weren’t training for. […]

When Klingberg’s study came out, both Jaeggi and Buschkuehl were doctoral candidates in cognitive psychology at the University of Bern, Switzerland. Since his high-school days as a Swiss national-champion rower, Buschkuehl had been interested in the degree to which skills — physical and mental — could be trained. Intrigued by Klingberg’s suggestion that training working memory could improve fluid intelligence, he showed the paper to Jaeggi, who was studying working memory with a test known as the N-back. “At that time there was pretty much no evidence whatsoever that you can train on one particular task and get transfer to another task that was totally different,” Jaeggi says. That is, while most skills improve with practice, the improvement is generally domain-specific: you don’t get better at Sudoku by doing crosswords. And fluid intelligence was not just another skill; it was the ultimate cognitive ability underlying all mental skills, and supposedly immune from the usual benefits of practice. To find that training on a working-memory task could result in an increase in fluid intelligence would be cognitive psychology’s equivalent of discovering particles traveling faster than light.

New York Times: Can You Make Yourself Smarter?

Hurley mentions one unpublished study that has failed to replicate the n-back results, but otherwise it is still holding up in tests. However, you should always be weary of the decline effect.

But really, the biggest drawback is probably that it’s hard to get people to start or stick with the n-back. I’ve known about for years now and still haven’t done it.