The benefit comes from that many "small" things you do in C# become much nicer in F#:
C#:
int target;
int someValue;
if (dict.TryGetValue("key", out someValue)) {
target = someValue * 100;
} else {
target = 42; // default
}
let target = match dict.TryGetValue "key" with
| true, x -> x * 100
| _ -> 42 // default
A more large scale win is any async code. In F#, you can use the async workflow (monad), and with 1 character (!), you make the compiler emit code that'd take at least 10 extra lines in C#. Inside an async workflow, a let! (example: let! foo = bar.AsyncGetResponse()) is like calling bar.BeginGetResponse, bar.EndGetResponse, while preserving the current exception handlers and all locals. It makes heavily async code trivial to write, where in C#, it's near impossible.
If it was on the JVM, looks like a serious competitor to Scala with some distinct advantages (! operator for sure) along with a little simpler syntax.
* Inbuilt support for sequence head and tail - very important for recursive functions.
* Due to the heavily object-oriented nature of C#, type inference is really limited. This makes writing anything non-trivial in functional style rather tedious as your type declarations will become several lines long really quickly. For example, try writing a Y Combinator in C# and F# to see the difference.
* Parallel processing is a lot easier in F#.
* Matching in F# is a lot more powerful than switch-case in C#, which makes functional programming easier.
My point is: functional programming support in C# is good enough for simple things but serious functional programming is a lot easier in F#. On the other hand, OO is a bit cumbersome in F#. I wouldn't use F# for typical web or enterprise create-update-delete database stuff but it is really great for things where functional programming is usually strong at, especially if you need it to work together with C# code.
There's nothing in functional programming that prevents it from using in "typical web" apps though, as demonstrated by the engine that runs HN, news.arc, for example.
No showstoppers, just minor annoyances that add up over time so when I write F# code I try to limit OO code to interfacing with other libraries (and the framework). Maybe it's just me, perhaps other programmers don't find OO in F# tedious at all.