Hi Chris,
- High-performance HTML combinators The first step would be measuring performance of various existing libraries Nothing is faster than caching (?) So maybe we can have faster pages writing cache strategies (?) Not using caching means less dev - time. So I agree on that this can be important.
Data storage - abstraction of relational databases (something like HaskellDB? based on Takusen?) - abstraction of nosql databases - purely functional databases (something like Happstack?)
No matter what you're going to do. It will be compared against the ease of use of use of existing RDBM solutions. You just connect to the database and run arbitrary queries. Also keep in mind that some projects require keeping track of values such as sums or max which is typically done by triggers. Haskell could do this as well.
- a good URL routing library (where you can only produce and link to valid urls) Think about this twice: You can't recompile bigger CMS systems each time you add a link. This may be fine for you personal web page (However you could use Joomla or such for that as well *kidding*)
No matter what you're going to do - Having kind of meta compiler (EDSL) in order to write Flash / JS code would be perfect - and maybe Haskell is the best language to write that. Have a look at the features of HaXe. It's interesting: Write one language and target Flash, JS, PHP, neko virtual machine, (IPhone).. Marc Weber