While I don't doubt Clojure 1.3 can get similar performance to Java, that's not what this benchmark shows. The Java and Clojure functions he's comparing don't even take and return the same argument types. He's comparing boxed Integers to primitive longs.
In the comments he says he saw no difference between Integer and ints. When I ran Tak.java with longs like the Clojure code my time went from 13s to 8s.
You're definitely right. I don't know why I received the very same results after having switched from Integer to int, I must have messed it up. Now it behaves with the same timings you describe. I'll rerun the tests and update the post accordingly.
Clojure is a language that has been teaching me a lot about different avenues of programming I hadn't previously encountered. I know many of them aren't new, but they are new to me and yet still approachable.
I really like the type-hinting approach; more detail and performance if you want or need, but otherwise don't bother. I guess it still doesn't have the type-safety of a static language, but now it has the performance.
I bet that something similar to the http://www.it.uu.se/research/group/hipe/papers/succ_types.pd... success typing used in erlang's dialyzer could be used to provide some level of type checking. Type hinting would only make this simpler (and with a strong contract).
In the comments he says he saw no difference between Integer and ints. When I ran Tak.java with longs like the Clojure code my time went from 13s to 8s.