Interesting discussion of inefficiencies in elisp bytecode during function calls: https://www.xemacs.org/Architecting-XEmacs/faster-elisp.html (From the XEmacs site, but entirely applicable to GNU Emacs and Remacs too I think)