Given a function with one fixed argument and a vararg, it should be sufficient to force evaluation of 2 elements for apply to know which arity it should select, however it currently forces 3:
This makes lazy functions that use apply (for example mapcat) less lazy than they could be.
The proposed patch makes RT.boundedLength short-circuit immediately after the seq count is greater than the max fixed arity:
The patch in this ticket slightly improves the issue reported at http://dev.clojure.org/jira/browse/CLJ-1218