If I had to pick one single advantage of F# over Haskell, it would be the general lack of partial functions. F# also does some really neat things with pattern matching. (See, for example, active patterns.)

The C#-style object model is arguably an advantage for certain purposes. I always feel guilty doing it, but object oriented domain modeling really does feel more natural for certain classes of problem.

On the downside, F# lacks higher kinded types, and you're absolutely on your own when it comes to making sure things that should be pure are actually pure.