I don't think you should rip out optimizations just because they don't
produce a clear signal above the noise of compile time. If you do a test
of just read, and it still makes no difference, then it would be more
reasonable to rip it out. If it still makes no difference it that test,
then it is likely that some bit decay has defeated this optimization, maybe
Gray streams support or something.
I will admit however, this this particular hack was added when CPUs were
thousands of times slower than they are today, and it may be time to
reevaluate. w.r.t. MP semantics, it might be reasonable to lock the stream
in the PREPARE-FOR-FAST-READ-CHAR.