ghcjs is a pretty heavyweight environment. It's actually translating the compiled core IR from GHC into JavaScript, and including a port of GHC's runtime system.

It can run basically any Haskell library. That includes the ability to run multithreaded Haskell code -- the JS RTS includes a scheduler. You also get Software Transactional Memory. Lazy evaluation works just as usual -- and so on.

The tradeoffs become worth it when you have a sufficiently valuable base of Haskell code that you want to run in the browser, and when your users aren't very constrained by page load time. Say, if you have some complicated tricky logic that you don't want to rewrite in another language, but you want to use it in your web app.