One problem with your code is that it scales as O(m * n), where  m == scalar @Array_1 and n == scalar @Array_2

Here's a solution that runs in O(m + n) instead, which should be much faster for large arrays:

use strict;
use warnings;
use 5.010; # only needed for say()

my @Array_1 = ("a1","a2","a3","a4","a5","a6");
my @Array_2 = ("a1","b2","c3","a4","f5","a6");
my @Array_3 = ("1","2","3","4","5","6");

my %seen;
@seen{@Array_2} = undef;
my @idx = grep exists $seen{$Array_1[$_]}, 0..$#Array_1;
@Array_1 = @Array_1[@idx];
@Array_3 = @Array_3[@idx];

say "@Array_1";
say "@Array_3";
[download]

There are several other ways to write that same algorithm (for example you could use splice to delete array elements one by one in-place, or push onto two new arrays in parallel), but only a benchmark shows which one is fastest.