I just wanted to point out that there other reasons than speed to use tail calls.

But after spotting and correcting the typo you had in in f0() it becomes more evident

use Benchmark qw(cmpthese);
use diagnostics;
 
 
my $limit = 10_000;
sub f0 { ++$_ < $limit ? &f0 : return $_ }
sub f1 { ++$_ < $limit ? goto &f1 : return $_ }
sub f2 {{ ++$_ < $limit ? redo : return $_ }}
cmpthese -1, {
   f0 => sub { $_ = 0; f0() },
   f1 => sub { $_ = 0; f1() },
   f2 => sub { $_ = 0; f2() },
};
[download]

Deep recursion on subroutine "main::f0" at /home/lanx/B/PL/PM/tail_cal
+l.pl line
    6 (#1)
    (W recursion) This subroutine has called itself (directly or indir
+ectly)
    100 times more than it has returned.  This probably indicates an
    infinite recursion, unless you're writing strange benchmark progra
+ms, in
    which case it indicates something else.
    
     Rate   f1   f0   f2
f1 76.1/s   -- -31% -73%
f0  111/s  46%   -- -60%
f2  280/s 267% 152%   --
[download]

Tail calls reuse the current call frame and thus avoid stack overflows in deep recursions.

I was able to make my system heavily swapping and unresponsive with a limit of 1000000.