Actually, thinking about it a bit more I think there is a natural explanation. If measured ability follows a normal distribution then the farther out you go the more spread out (on average) individuals become. For example, draw a billion samples from the right half of a normal distribution with mean 0 and stddev 1. There will be more drawn between 4 and 5 than between 5 and 6. The extreme outliers will be in some sense more lonely.
Edit: here are the most extreme values from a run of a billion draw with the distance to the next smallest value.
"Ability" isn't a single variable either. There's a combination between natural talent and hard, determined work. For instance, many people thought Ricardo Quaresma and Cristiano Ronaldo had similar amounts of natural talent as young players, but Quaresma wasn't as determined as Ronaldo to become the greatest player in the world. Conversely, Michael Jordan's contemporary Charles Barkley didn't really have Jordan's natural talent, but made up for it by working exceptionally hard. So you have at least two (and possibly more) variables, all of which have to reach extreme outliers together, to produce an athlete like Messi, Jordan, or Gretzky.
Edit: here are the most extreme values from a run of a billion draw with the distance to the next smallest value.
5.967211966 0.056293894
5.910918072 0.043174783
5.867743289 0.081026984
5.786716305 0.043416671
5.743299634 0.094533593
5.648766041 0.000088773
5.648737164 0.005932669
5.642804495 0.011669773
5.631134722 0.003206244
5.627928478 0.004643796
5.623284683 0.013723654
5.609561028 0.014653298
5.59490773 0.018714748
5.576192982 0.001690759
5.574502223 0.031026791
5.543475432 0.002331827
5.541143605 0.022237834
5.51890577 0.004242771
5.514662999 0.009347961
5.505315038 0.004298492
5.501016547 0.02271157