I see the opposite. Where's your test? I never trust a benchmark I don't see; they're too easy to get wrong.

>perl a.pl
          Rate test2 test1
test2 225043/s    --   -5%
test1 236071/s    5%    --

>perl a.pl
          Rate test2 test1
test2 233861/s    --   -5%
test1 247248/s    6%    --

>perl a.pl
          Rate test2 test1
test2 224170/s    --   -8%
test1 243265/s    9%    --
[download]

$ perl a.pl
          Rate test2 test1
test2 194855/s    --  -10%
test1 217061/s   11%    --

$ perl a.pl
          Rate test2 test1
test2 200313/s    --   -5%
test1 210977/s    5%    --

$ perl a.pl
          Rate test2 test1
test2 195321/s    --   -4%
test1 203538/s    4%    --
[download]

use strict;
use warnings;

use Benchmark qw( cmpthese );

sub test1 { my %h = 'a'..'z'; %h = (); }
sub test2 { my %h = 'a'..'z'; undef %h; }

cmpthese(-3, {
   test1 => \&test1,
   test2 => \&test2,
});
[download]