Reflection is generally the #1 reason for slow Clojure code. Especially when used inside loops.
Other things to keep in mind sometimes is the use of checked math, for math heavy code, disabling the checks for overflows can help speed things up.
The fact that some reflection doesn’t show up with warn-on-reflection, specifically aget and aset don’t, and their multi-arity variant uses reflection under the hood making them super slow.
Then using sequences can slow things a bit if a lot of intermediate data is created, moving to transducers can help speed it up. Or loop/recur for utmost performance.
When using persistent data-structures and doing a lot of modification, transient can help speed things up. And some problems that really need fast updates or inserts/deletes can benefit from switching to arrays or using Java’s mutable data-structures instead.
Finally, primitive calls to loop/recur and functions can help speed boxing overhead in code that does a lot of primitive data manipulation.
Oh and I guess there’s also the use of protocols to speed up polymorphism if that’s an issue.
Anyways, for all these, I’d only reach for them if I notice slowness. And I like to use a profiler to identify where to improve things, such as https://github.com/clojure-goes-fast/clj-async-profiler or Visual VM