The _sensible approach seems the most sane way to me. That said, there are a few things to "fix".

First - what are the &'s doing there? Experienced perl programmer? As I said before, seeing that & is an indicator of low experience, not high experience, to me. It's an indicator of perl4 programming, which, granted, is a long time programming in perl, but does not equate to high amounts of good experience in modern perls. Unless, of course, it were used in a goto, or magic argument passing, or overriding prototypes. (Of course, the use of prototypes at all can be suspect.)

The first two options both take a parameter. Then ignore it. Your version uses your parameter. That's already a plus for your version.

And the sensible version clobbers $_. Bad. Maybe he meant to say:

  local $_ = shift;
[download]

I can buy that. Except that he's not really using $_ for anything that $t wouldn't work for as a lexical variable.

Finally, the subtraction is a bit funny. You get a funny lookin' number for the -1 time (-1 day plus 23 hours plus 59 minutes plus 59 seconds - technically correct, but may not be what you want). Perhaps what he meant was "$_ = int($_ / 60)". In that case, here's a slightly revamped sensible function:

sub sec_to_dhms_sensible2
{
    use integer;
    local $_ = shift;
    my ($d, $h, $m, $s);
    $s = $_ % 60; $_ /= 60;
    $m = $_ % 60; $_ /= 60;
    $h = $_ % 24; $_ /= 24;
    $d = $_;
    return ($d, $h, $m, $s);
}
[download]

I didn't fix everything, but it's a bit cleaner to look at. The only difference in output is that this gives a seconds of -1 rather than -1d, +23:59:59. Note that the use integer; is block-scoped (I confirmed this in testing) so that it doesn't affect anything else in the file - you can use floating point elsewhere.