use Benchmark;
my %hash;
my ($c, $i) = 1000, 100;
for (1 .. $c) {
    $hash{"test$_"} = $_;
    $hash{"foo$_"} = $_;
}
sub SUBSTR {
    foreach my $bid (sort {
        substr ($a,4) <=> substr ($b,4)
    } grep {/^test(\d+)$/} keys %hash) {
    }
}
sub SCHWARTZ {
    for my $bid (map { $_->[1] } sort {
        $a->[0] <=> $b->[0]
    } map { m/^test(\d+)$/ ? [$1, $_] : () } keys %hash) {
    }
}
timethese($i, { SUBSTR => \&SUBSTR, SCHWARTZ => \&SCHWARTZ });

Benchmark: 
timing 100 iterations of
 SCHWARTZ, SUBSTR
...

  SCHWARTZ:  1 wallclock secs ( 0.77 usr +  0.00 sys =  0.77 CPU) @ 129.87/s (n=100)

    SUBSTR:  1 wallclock secs ( 0.91 usr +  0.00 sys =  0.91 CPU) @ 109.89/s (n=100)