OP's concern was not the use of keys, but the potential cost of using it due to a perceived 'temporary array'. However various solutions to that problem have been suggested. Let's see how yours stacks up:

use strict;
use warnings;
use Benchmark qw(cmpthese);

my %bigHash   = map {$_ => undef} 1 .. 1e3;
my ($key)     = each %bigHash;
my $count     = keys %bigHash;
my ($key2)    = (%bigHash)[0];

print "First key '$key', second '$key2'\n";

cmpthese (
    -1,
    {
        byEach  => sub {scalar keys %bigHash; my ($key) = each %bigHas
+h;},
        byArray => sub {my $key = (%bigHash)[0]},
    }
);
[download]

Prints:

First key '559', second '559'
             Rate byArray  byEach
byArray    5228/s      --   -100%
byEach  2414187/s  46082%      --
[download]

It's that 'temporary array' OP was worried about. A fancy solution no doubt but, for saving time, not a good one.