This made me wonder if the choice of which side to castle is consistent between chess960 and traditional chess. In other words, after castling do the Kings end up on the two sides of the board -- Kingside or Queenside (h-side or a-side in chess960 parlance) -- with the same frequency in both versions of the game. The game scores from the 2009 Chess960 Rapid World Championship (RWC), won by Nakamura, seemed to indicate that there was no castling in many games and that castling O-O-O occurred more often castling O-O. This is contrary to my experience in traditional chess and I decided to do a simple analysis.
First I looked at the games from the recent Pearl Spring tournament, won by Carlsen. In the 30 games played there I found that O-O occurred 42 times and O-O-O occurred 14 times. This means that of the 60 times the castling option was presented (30 games x 2 players per game), the players castled 56 times. In those games where the players castled, castling O-O occurred 75% of the time and castling O-O-O 25%. These results were approximately what I expected to see.
The 2009 RWC presented a different picture. In the 20 games played, O-O occurred 9 times and O-O-O occurred 14 times. This means that of the 40 castling opportunities, the players castled only 23 times, and O-O-O was played more often than O-O.
Would previous years confirm this result? The year 2008 saw a match for the Women's RWC title, with fewer games than in 2009, so I skipped to 2007. The 2007 RWC featured 22 games. The move O-O occurred 29 times and O-O-O occurred 8 times. These results were more in line with the Pearl Spring statistics.
My conclusion was that the samples were too small to be meaningful. I'll come back to this question another time when I have more data to analyze.