So basically you measured startup time. The same without startup time:

#!/usr/bin/perl
use strict;
use warnings;


package NumberHolder;
use Moose;

has 'x' => (is => 'rw', isa => 'Int');
has 'y' => (is => 'rw', isa => 'Int');

__PACKAGE__->meta->make_immutable;

package NumberRun;

sub new {
        my $class = shift;
        my %args = @_;
        my %h = ( "x" => 0, "y" => 0);
        @h{ keys %args} = values %args;
        return bless \%h;
}

sub x {
        my $self = shift;
        $self->{x} = shift if @_;
        $self->{x};
}

sub y {
        my $self = shift;
        $self->{y} = shift if @_;
        $self->{y};
}



package Main;

use Benchmark qw( cmpthese );

sub withMoose {
    my $z = NumberHolder->new("x" => 1, "y" => 2);

    my $f= $z->x()+ $z->y(); 

    if ($z->x()>0) {
    $z->y(5);
    }
}

sub traditional {
    my $z = NumberRun->new("x" => 1, "y" => 2);

    my $f= $z->x()+ $z->y(); 

    if ($z->x()>0) {
    $z->y(5);
    }
}




my %tests = (
   "with Moose"    => sub { withMoose  },
   "traditional"    => sub { traditional },
);


cmpthese(-5, \%tests);

# prints:
               Rate  with Moose traditional
with Moose   49776/s          --        -55%
traditional 111550/s        124%          --
[download]

I eliminated the prints as these take a lot of time themselves. Put them in again and the difference will be 80% instead of 124%.

Now this represents worst case. The methods are minimalist and the test subs just call the methods a few times. Print, read or write to files or use a database and the difference gets smaller and smaller