Actually, Graham worked with the CLISP implementation, which is not exactly noted for being the fastest one around. It's quite possible that Yahoo would have been able to get the results they wanted speedwise by switching to a more suitable implementation of Common Lisp. (For example, type and optimization declarations
are ignored, even now, by CLISP.) And while Graham is obviously a very good and respected Lisp programmer (I think I'd stop short of calling Mr. Lisp, though), even he's going to have a problem working around something like that. It would have been interesting to see how it fared if Yahoo had chosen instead to port it to something like Allegro CL, Lispworks, or CMUCL.