A regex solution is significantly faster than a ST one.

Some benchmarks:

use warnings;
use strict;

use Benchmark qw(cmpthese);


my $multiplier = 10_000;
my $s__ = "abcdefghij" x $multiplier;  # base string
my $s10 = "Abcdefghij" x $multiplier;  # 10% differ
my $s50 = "ABCDEfghij" x $multiplier;  # 50% differ
my $s90 = "ABCDEFGHIj" x $multiplier;  # 90% differ


sub differ_at_st {
    my $i = 0;
    return [
        map   $_->[1],
        grep  $_->[0] ne qq{\x00},
        map   [ $_, $i++ ],
        split '',
        $_[0] ^ $_[1]
        ];
    }

sub differ_at_rx {
    my $diff = $_[0] ^ $_[1];
    my @diffs;
    push @diffs, $-[1] while $diff =~ m{ ([^\000]) }xmsg;
    return \@diffs;
    }


print "multiplier == $multiplier \n";
cmpthese(-10, {
    st_10 => sub { differ_at_st($s__, $s10) },
    st_50 => sub { differ_at_st($s__, $s50) },
    st_90 => sub { differ_at_st($s__, $s90) },
    rx_10 => sub { differ_at_rx($s__, $s10) },
    rx_50 => sub { differ_at_rx($s__, $s50) },
    rx_90 => sub { differ_at_rx($s__, $s90) },
    });
[download]

Output:

C:\@Work\Perl\monks\745091>perl 745091_1.pl
multiplier == 1000
        Rate st_90 st_50 st_10 rx_90 rx_50 rx_10
st_90 10.7/s    --   -0%  -12%  -49%  -73%  -94%
st_50 10.7/s    0%    --  -11%  -49%  -73%  -94%
st_10 12.1/s   13%   13%    --  -42%  -70%  -93%
rx_90 20.9/s   95%   95%   73%    --  -48%  -88%
rx_50 40.0/s  274%  273%  230%   91%    --  -77%
rx_10  177/s 1562% 1555% 1367%  750%  344%    --

C:\@Work\Perl\monks\745091>perl 745091_1.pl
multiplier == 10000
        Rate st_90 st_50 st_10 rx_90 rx_50 rx_10
st_90 1.03/s    --   -2%  -11%  -50%  -73%  -94%
st_50 1.05/s    2%    --   -9%  -49%  -72%  -94%
st_10 1.15/s   13%   10%    --  -44%  -69%  -93%
rx_90 2.05/s  100%   96%   78%    --  -45%  -88%
rx_50 3.76/s  267%  260%  226%   83%    --  -77%
rx_10 16.6/s 1513% 1483% 1334%  707%  340%    --
[download]