Following the "The Emacs master is much slower than the emacs-27 branch." 
thread, I tested Alan's benchmark on the master branch.  The platform I 
used is a 64-bit GNU/Linux computer, with GCC 10.2.

It turns out that the 10-15% slowdown of redisplay in Emacs 28, compared 
to Emacs 27, is due to a single character change, namely the "!" character 
removed in commit 165fd028.

I attach three graphs of the timings calculated on various revisions of 
master.  The benchmark ran ten times on each revision; Emacs was compiled 
with CFLAGS='-O2'.  "timing-1.png" shows that the slowdown happens between 
revision 68ff32a5 and 7445560d, "timing-2.png" narrows this to revisions 
4c5043c5 to d3c73fbf, and "timing-3.png" identifies revision 165fd028 as 
the cause of the problem.