Oh, I love these kind of Unicode bugs. Also, please don't use utf16 in C# - although it advertises as using utf16, all of it's built-in string processing actually works on code units, not Unicode code points and breaks down on surrogate pairs.

Anyway, each time you start a new project that is supposed to work correctly with multi-language data, please set up evil strings test suite as early as you can and do not assume that any of the 3rd party software that you're using won't break down.