Well, tr appears to win if the 'M' is a common occurrence.

#! /usr/bin/perl -lw

sub cnt_a {
  my ($count, $str) = (0, shift);
  ++$count while ($str =~ /(M+)/g);
  $count;
}

sub cnt_b {
  my ($count, $str) = (0, shift);
  ++$count while ($str =~ /M+/g);
  $count;
}

sub cnt_c { int (() = shift =~ /M+/g) }

sub cnt_y { shift =~ y///csr =~ y/M// }


use Benchmark qw(cmpthese);

my @S = qw(I M O X);
# my @S = map chr, 0..255;
our $str = pack "(a)*", map $S[rand @S], 1..100000;

print for map $_->($str),
        qw(CORE::length cnt_a cnt_b cnt_c cnt_y);

cmpthese -3, {
  a => 'cnt_a($str)',
  b => 'cnt_b($str)',
  c => 'cnt_c($str)',
  y => 'cnt_y($str)',
}
[download]