if I understand the OP correctly,

Hence the reason for reserving the right to change my mind. I couldn't make head nor tails of this:

I get 1 seconds per 50 iterations. Obviously better than the 28 seconds, but still significantly higher than the 9 seconds

It also doesn't ring true that

The function itself is quite simple - it gets 3 scalars (one of which is a field of some object), checks a couple of if's on the values and some basic math (including modulo %), and returns an array with a couple of hashes, each with two numerical fields. That's it.

Would require 28 seconds/50 iterations.

A tentative attempt to match the description yields :

sub x{ 
    my( $s1, $s2, $s3 ) = @_;
    $s1 *= 10 if $s1;
    $s2 **= 4 if $s2;
    $s3 *= $s2 % $s1;
    return [ 
        { n1 => $s1, n2 => $s2 }, 
        { n1 => 41 * $s2, n2 => ( $s1 * $s2 ) % $s3 } 
    ];
};;
$I = 50e3; 
$t = time; 
x( 123, 456, 789 ) for 1 .. $I; 
printf "$I calls took %.4f seconds\n", time()-$t;;
50000 calls took 0.3174 seconds

$I = 50e5; 
$t = time; 
x( 123, 456, 789 ) for 1 .. $I; 
printf "$I calls took %.4f seconds\n", time()-$t;;
5000000 calls took 26.5590 seconds
[download]

I appreciate my sub is guesswork and nothing like the real thing, but given the description, it is hard to guess what else that description is hiding that causes the sub to take a million times longer?

Of course, it'll turn out that he's tallying the national debt. Or crowd sourcing the math to Twitter.