The problem states that performance matters, so I think the top spot should be the most readable solution with a O(n) complexity. I'm not saying mine should be there ;) As for the bug, I think you're being a little hard on me. The "bug" requires a string with at least 2GB of the same character (if int is 4 bytes, depends on the platform).

Edit: I've "fixed" my solution by using string's size_type. It is slightly less readable though because now the type is unsigned.

This solution is very idomatic and easily read. So it deserves it's spot at the top.

Your solution has a hidden assumed limit and a bug on the ammount of any specific symbol appearing in the string.
If s1 contains MAX_INT or more appearances of a single character your solution starts returning the wrong output.

It is disappointing that this solution is ranked first. While elegant, its complexity of O(nlogn) is higher than needed. Using a histogram-based solution is O(n).