Many Lisp implementations are in the speed range of SBCL. When you look at such an application, the super optimized part of the code is not that large. Equally, if not more, important is the robustness of the runtime system against all kinds of errors, the speed and responsiveness of the memory management system and the speed of compiled code which runs with full safety checks. Also if a Lisp system would use much of CLOS, the speed of the CLOS implementation and the library one writes with CLOS becomes important.

If there are speed problems in an application like Stumpwm with the current Lisp systems, it is definitely not the raw micro-bench speed that's a problem, but more likely something else (architecture, threading, GC, event handling, bugs, ...).