One more way. Copy your hash keys to arrays, and do a binary search. The setup is O(n*log(n)), but the advantage is that the lookups are O(log(n)) (Not quite the O(1) as you wanted, but pretty good). (all untested, but pretty much lifted from Array::IntSpan, which I've seen alot of lately :)

my @array = sort keys %hash;
# Keep another array for regexps
my @regexes = map qr/^$_/, @array;

for my $ph_num (@phone_numbers) {
 my $num = search($ph_num) || 'not';
 print "$ph_num $num found\n"
}

sub search {
 my $num = shift;
 my ($start, $end) = (0, $#array);
 while ($start < $end) {
    my $mid = ($start+$end)/2;
    if ($array[$mid] le $num) {
        return $hash{$array[$mid]} if $num =~ $regexes[$mid];
        $start = $mid+1;
    } else {
        $end = $mid-1;
    }
  }
  $num =~ $regexes[$start] && $hash{$array[$start]};
}
[download]

Update: Added slight optimization. We are still occasionally repeating the last regex, that's ok though. I don't think there's an optimal way to eliminate it (or I just don't feel liking looking for it at the moment :).