The NY Times has an article "Secrets of a Mind-Gamer". The journalist, Joshua Foer, who wrote also a book ("Moonwalking With Einstein: The Art and Science of Remembering Everything") gets a training in memory and takes part in a US memory competition.
First lesson: everybody can do it and most of the people who take part in memory competitions have an average intelligence. What makes the difference is a number of tricks or techniques. The most important of those is memory palaces where you associate every item with a known person in a strange act that is placed in some building that you know well. It can be any type of space as long as you know it well. On fMRI images mental athletes use the same parts of the brain as "normal" people with one exception: spatial memory.
In 1978, he and a fellow psychologist named Bill Chase conducted what became a classic experiment on a Carnegie Mellon undergraduate student, who was immortalized as S.F. in the literature. Chase and Ericsson paid S.F. to spend several hours a week in their lab taking a simple memory test again and again. S.F. sat in a chair and tried to remember as many numbers as possible as they were read off at the rate of one per second. At the outset, he could hold only about seven digits at a time in his head. When the experiment wrapped up — two years and 250 mind-numbing hours later — S.F. had increased his ability to remember numbers by a factor of 10.
In a famous experiment carried out in the 1970s, researchers asked subjects to look at 10,000 images just once and for just five seconds each. (It took five days to perform the test.) Afterward, when they showed the subjects pairs of pictures — one they looked at before and one they hadn’t — they found that people were able to remember more than 80 percent of what they had seen.
The author trained at remembering a pack of cards. He had to train every morning 10 to 15 minutes. He associated every card with an image that he placed in a building. After some time he stuck at one card per 10 second. His teacher compares this to the plateau top sporters and typists reach when their act is totally automated. He mentions two techniques to overcome this barrier. One is going deliberately faster - accepting that that results in regular errors - and then working on those errors. The other is spending most of your training time on what you can't do yet.