I find the Python ORM options (sqlalchemy/Django/etc) lot better evolved than what is available for clojure and couldn't image that being a roadblock. Korma is basically a query builder and lacks a lot of things I was used to, and I had to engineer a declarative layer for defining models in the ways I wanted.
Pain in setting up developer machines could be simplified with an ansible playbook pretty easily.
OTOH, one of the worst things I hate is trying to figure out how to automatically get around Sun's paywalls and fiddle with Java classpaths :)
I do feel some level of want to avoid vagrant in the development workflow, in which cases I'm usually fond of just logging into a real VM. I do think it's potentially good for tests.
IDK, I always want to solve the idea of "fragile deploys" with immutable systems in prod. Uberjar was cool, I'll give it that -- but the startup time of clojure was often not worth the wait.
My biggest complaints with clojure is documentation quality, inconsistency in return types (or just willingness to return nil in error scenarios), and tracebacks that are very difficult to sort through.
Clojure also presents some problems with the idiomatic nature of trying to roll up everything into one statement, whereas adding debug and debugging require unrolling it, than packing statements up again to be socially acceptable idiomatic clojure.
Testing in clojure also strikes me as painful, as it tends to punish you with "let" causing an indentation every time you need to assign a variable, something that is useful in tests that want to check results a lot along the way.
My gut feeling is clojure was good for a little light scripting on top of a big Java codebase (kind of like you might use Lua in a game engine) but would not be something I'd pick for a primary language. Primarily using Java libs would also solve some of the ecosystem problems.