I was curious, so...

use Inline C;

sub cmp_perl
{
    my ($a, $b) = @_;
    my $c = $a ^ $b;
    my @ret;
    push @ret, pos($c) while $c =~ /[^\0]/g;
    @ret;
}

use Benchmark ':all';

my ($x, $y);
$y = $x = 'A' x 1e6;
substr($y, $_*100, 1) = 'C' for 1..(1e4-1);
timethese(0, {
    perl => sub { cmp_perl($x, $y); },
    C => sub { cmp_c($x, $y); },
});
__END__
__C__
void cmp_c(char *a, char *b)
{
    int i, n;
    Inline_Stack_Vars;
    Inline_Stack_Reset;
    for (i=0, n=0; *a && *b; a++, b++, i++)
        if (*a != *b)
            mXPUSHi(i), n++;
    Inline_Stack_Return(n);
}
[download]

I was disappointed that the Perl version was about 8x slower:

Benchmark: running C, perl for at least 3 CPU seconds...
         C:  3 wallclock secs ( 3.17 usr +  0.00 sys =  3.17 CPU) @ 37
+9.18/s (n=1202)
      perl:  3 wallclock secs ( 3.17 usr +  0.00 sys =  3.17 CPU) @ 43
+.22/s (n=137)
[download]