With the files at 500MB each, unless you have a machine with a lot of memory, thats a lot to read into a hash.

If the rows are in any given order you could exploit that. If not then if you first sorted them with

tail +2 d1.dat > d1.sorted
tail +2 d2.dat > d2.sorted
[download]

The tail +2 just removes the header line from each. You can then merge them with

open(D1,"<d1.sorted");
open(D2,"<d2.sorted");

my $d1 = "v1 v2 v3";
my $d2 = "t1 t2 t3";
my ($d1k,$d2k)=('','');

for(;;) {
  my $c = $d1k cmp $d2k
    or print join(" ", split(/\s+/,$d1), (split(/\s+/, $d2))[2]),"\n";

  if ($c <= 0) {
    last unless defined($d1 = <D1>);
    $d1k = ($d1 =~ /^(\d+\s+\d+)/)[0];
  }

  if ($c >= 0) {
    last unless defined($d2 = <D2>);
    $d2k = ($d2 =~ /^(\d+\s+\d+)/)[0];
  }
}
[download]