It works well in that benchmark, but falls down here:

use Benchmark qw(cmpthese);

my $str = "wwibblewibblewibblewibbleibblewibblewibblewibble";

cmpthese -1, {
    regex => sub {
        my %count;
        ++$count{$_} for $str =~ /(?=(..))/g;
    },
    substr => sub {
        my %count;
        ++$count{substr $str, $_, 2} for (0 .. length($str) - 2);
    },
    unpack => sub {
        my %count;
        my $n = length($str) - 1;
        ++$count{$_} for unpack qq{(a2 X)$n}, $str;
    },
};

          Rate  regex unpack substr
regex  15316/s     --   -43%   -74%
unpack 26935/s    76%     --   -54%
substr 58514/s   282%   117%     --
[download]

substr slows down 50% if the string contains utf-8 characters, but it's still significantly faster than unpack in this benchmark:

my $str = "wwibblewibblewibblewibbleibblewibblewibblewibble\x{20ac}";

          Rate  regex unpack substr
regex  14222/s     --   -35%   -51%
unpack 21976/s    55%     --   -24%
substr 29020/s   104%    32%     --
[download]