just use sort to order the keys of the hash

As a general rule, don't sort to retrieve the min/max of a list. That's spectacularly inefficient. Scan the list once for a cost of O(n) and retrieve the information required. If you sort, depending on your algorithm and dataset you'll usually get around O(n log n). A quick benchmark bears this out:

#! /usr/local/bin/perl -w

use strict;
use Benchmark qw/cmpthese/;

my %h;

$h{ int(rand 1_000_000) } = rand 1000 for 1..10_000;

sub min_by_sort { ( sort {$a <=> $b} keys %h )[0] }
sub min_by_scan {
    my $min = undef;
    (not defined $min or $min > $_) and $min = $_ for keys %h;
    $min;
}

print "min_by_sort = ", min_by_sort(), "\n";
print "min_by_scan = ", min_by_scan(), "\n";

cmpthese(
    shift || -5,
    {
        min_by_sort => \&min_by_sort,
        min_by_scan => \&min_by_scan,
    }
)
[download]

For a list of ten elements, the scan does much better than the sort:

min_by_sort = 107
min_by_scan = 107
Benchmark: running min_by_scan, min_by_sort, each for at least 5 CPU s
+econds...
min_by_scan:  4 wallclock secs ( 5.44 usr +  0.00 sys =  5.44 CPU) @ 9
+5284.97/s (n=518112)
min_by_sort:  6 wallclock secs ( 5.12 usr +  0.00 sys =  5.12 CPU) @ 1
+52355.18/s (n=779630)
                Rate min_by_scan min_by_sort
min_by_scan  95285/s          --        -37%
min_by_sort 152355/s         60%          --
[download]

update: eserte points out that I misread the results. Somehow I managed to past in the results of something else (and I forget what). The correct results for a list of 10 elements are as follows:

min_by_scan:  6 wallclock secs ( 5.30 usr +  0.00 sys =  5.30 CPU) @ 9
+3681.46/s (n=496219)
min_by_sort:  6 wallclock secs ( 5.28 usr +  0.00 sys =  5.28 CPU) @ 7
+1843.03/s (n=379421)
               Rate min_by_sort min_by_scan
min_by_sort 71843/s          --        -23%
min_by_scan 93681/s         30%          --
[download]

If the hash contains around 10 000 keys (note that the quick and dirty initialisation used means that I might generate the same key twice) the scan method is still a winner, although sort is catching up because of the pure-C/perl-ops diffential.

min_by_sort = 301
min_by_scan = 301
Benchmark: running min_by_scan, min_by_sort, each for at least 5 CPU s
+econds...
min_by_scan:  6 wallclock secs ( 5.48 usr +  0.01 sys =  5.48 CPU) @ 9
+4.63/s (n=519)
min_by_sort:  6 wallclock secs ( 5.30 usr +  0.00 sys =  5.30 CPU) @ 7
+3.14/s (n=388)
              Rate min_by_sort min_by_scan
min_by_sort 73.1/s          --        -23%
min_by_scan 94.6/s         29%          --
[download]

So for really large hashes you should use sort, otherwise scan. This is only due to the fact that sort is a single perl op (modern perls can recognise sort {$a <=> $b}) at which point you drop down into optimised C. The scan loop will run a small loop of ops through the interpreter which limits its flat-out performance. If you were writing both in C the scan would always win.

For instance, if you disable the inlined sort by creating a comparator like sort {$a+1 <=> $b+1} sort's improvement quickly evaporates and the scan continues to be the best performer. It's something peculiar to Perl, and something to keep in mind.

update: A thought just came to me over lunch: even if you consider the speed argument weak (which is fine by me) a more damning problem is how much memory is used. The sort will require a second copy of the list to be held in memory. Given Perl's propensity to consume memory, this will be a significant factor for large lists.