I've seen discussion about eval performance on Clojure mailing list and decided to test the speed of Clojure's eval, additionally to other Lisp dialects I tested previously. It turned that Clojure has slower eval than other Lisp implementations. Althought Clojure has two features that slow down eval - macros like most of Lisp dialects and some of the built-in primitives are macros as well, as in CL, I guess that the main reason for slow eval in that particular test is in integration with Java.
All results and the code are integrated in my older post:
Speed of Eval in Some Lisp Implementations.