Play Framework vs. the relative messiness of Yesod/Snap/Happstack/Scotty/raw WAI. Clojure has a similar (to Haskell's) story here. I'm understating it here, but this is pretty big. Most web applications are not special snowflakes, a framework to wrap-up common-case nonsense concisely is very valuable. Micro-framework stacks like using WAI directly or Scotty are good to have, but insufficient.
Specialized NLP libraries (in Java, at present) such as for medical lexicons. Might be replicable by being layered on top of existing Haskell NLP libs, but doesn't yet exist to my knowledge. Hardening up libraries like this usually takes a lot of time.
Slick is nicer and easier to get started with than Persistent. Partly due to documentation, partly because Slick isn't silly and embraces relational databases directly, partly because Persistent is just kind of hairy. Supporting MongoDB isn't a plus, it's a minus. Compromises the API anyway.
A migrations library of pretty much any sort at all.
Scala/xml equivalent in Haskell would be nice.
Rather critically for me personally, I need a Haskell DSL client library for Elasticsearch like this Scala one: https://github.com/sksamuel/elastic4s
This is the only front on which I'm likely to be able to personally contribute due to my past experience writing database client libraries and working with Elasticsearch. Haskell doesn't have a working, let alone documented and nice, ES client library of any kind. This is bad.
Of particular note are the websites and documentation associated with these projects:
This stuff matters! A Hakyll template for a standard Haskell library documentation site might be useful here.
I need to write a Haskell version of Scala School: http://twitter.github.io/scala_school/ - probably typeclass and typeclassopedia-centric to Sterling's eternal chagrin. However, there's a time constraint and the goal is to get people rolling quickly.
Steal everything nice that Scala has and give nothing back.
Can I file it under "when Scala does better than Haskell" ? :)