Hmm. Works for me?

#! perl -slw
use strict;
use Benchmark qw[ cmpthese ];


our $str = 'azbzczdzezfzgzhzizjzkzlzmznzozpzqzrzsztzuzvzwzxzyzzz';
our $SrchStr = 'z';

my $n1 = eval "\$str =~ tr/$SrchStr/$SrchStr/";
my $n2 = ( $str ^ ($SrchStr x length $str ) ) =~ tr[\0][\0];
my $n3 =  @{[ $str =~ m/$SrchStr/g ]};
print "$n1 : $n2 : $n3";

cmpthese -1, {
    eval => q[ my $n = eval "\$str =~ tr/$SrchStr/$SrchStr/" ],
    xor  => q[ my $n = ( $str ^ ($SrchStr x length $str ) ) =~ tr[\0][
+\0]],
    match=> q[ my $n =  @{[ $str =~ m/$SrchStr/g ]} ],
};

__END__
[14:50:30.01] P:\test>eval-v-xor.pl
27 : 27 : 27
           Rate  eval match   xor
eval    24183/s    --  -23%  -98%
match   31450/s   30%    --  -97%
xor   1127154/s 4561% 3484%    --
[download]