For one, Haskell actually has a pretty good and growing collection of libraries, probably disproportionate to the language's popularity. Web development, in particular, is well represented: there are several competing web frameworks; Yesod even has commercial backing now. Haskell even has some libraries that are not well reproduced in other languages like QuickCheck and Parsec. I've also been looking at sbv lately, which seems rather more convenient than other languages' alternatives.
Additionally, libraries aren't the only thing that matters. In fact, most libraries are pretty trivial. I remember seeing a presentation by Brian O'Sullivan online about his experiences with using Haskell at a startup. One point that really stood out to me was about libraries: sure, some libraries (like JSON parsing, I believe) were missing; however, it only took him something like a couple of days to write it from scratch. Those days are easily overshadowed by Haskell's core advantages. I would much rather spend a couple of days writing a new library than trying to track down a bug Haskell's type system would have caught. The beauty of this system is also that each library only needs to be written once and can then be reused, making the world better for everyone.
Now, there are some other issues with Haskell. Cabal really is a big one. I just spent quite a while fighting cabal, and I'm not even sure how I won in the end :P. Happily, I believe people are working on addressing the cabal issue.
I'm sure there are other annoyances besides. But no language is entirely free of annoyances. Considering everything, I really do think Haskell is a good choice for a very wide range of programming problems, especially in a startup setting.