From the discussion of their most relevant benchmark (Singular value decomposition):
The Allegro CL 4.1 times of 3.9 seconds beat the f77 time of 4.8 [seconds];
Nice!
setting on optimization for f77 brought its time down to 0.45 seconds.
Thus for this system, the [LISP] compiled code can have quite comparable speed to
that of the corresponding unoptimized Fortran in this case as well.
It's been a few decades since I have read the paper, so it seems my memory is a bit fuzzy on that regard.
On the other hand, given that C always had issues to beat even unoptimized Fortran, due to the optimization restrictions before C99, it is quite commendable that Lisp achieves such results.
It's not obvious how optimized C in the '80s wouldn't have been as fast as unoptimized Fortran 77, even with whatever optimization restriction you might be thinking of.
The way the authors of that paper talk about unoptimized code in that paper gives the impression that they don't know what they're talking about. Your comments here begin to put you at risk of a similar appearance.