I don't know if there is a specific issue with Solaris or Hpux, but it's not really a bug in general. You just need to understand that the mainwindow, IS the event-loop, and if you destroy the event-loop, all processing stops. This keeps the event-loop going. Does this work?

#!/usr/bin/perl -w
use Tk;

#the leakage stops after about 10 open-close cycles

#read perldoc Tk::callbacks
#this is a better method than the first example

my $mw = MainWindow->new(-title=>"Demo");

my $HlpBttn = $mw->Button(
   -text    =>"Make NEW Window",
   -command =>  [ \&makeWindow, $mw ], 
   # calls makeWindow and passes $mw as the first argument
);
                     
$HlpBttn->pack(-padx =>60, -pady =>10);


MainLoop;

sub makeWindow {
   my $parent = shift;
   my $win  = $parent->Toplevel(-title=>'new window');
   my $Bttn = $win->Button(
       -text=>"CLOSE this window",
       -command=>  [ 'destroy', $win ],  # calls destroy on $win
   )->pack(
       -padx =>60,
       -pady =>10,
   );
}
__END__
[download]

Here is another script which demonstrates continuation after mainloop

#!/usr/bin/perl -w
use Tk;

my $mw = MainWindow->new(-title=>"#1");

$mw->Button(-text=>"Destroy Me", 
         -command=> sub { $mw->destroy })->pack;
                 
MainLoop;

print "ha ha still going\n";
<>;
[download]

So if you want to exit the program use sub{ exit; }

exit if defined $mainwindow
[download]

Of course, I'm speaking in generalities, there may very well be some oddity on your system. If you give a working snippet, we can report what we see.