Now, I wonder about Python vs Perl. And then, something wonderful...

Python

from sys import stdout
from multiprocessing import Process
from multiprocessing import Barrier

def task(barrier, number):

    for i in range(1, 400 + 1):
        print(f'{i}: {number}')
        barrier.wait()

if __name__ == '__main__':

    stdout.flush()
    num_workers = 500
    workers = list()

    # create a barrier
    barrier = Barrier(num_workers)

    # create the worker processes
    for i in range(1, num_workers + 1):
        workers.append(Process(target=task, args=(barrier, i)))
        workers[-1].start()

    # wait for all processes to finish
    for w in workers:
        w.join()
[download]

Perl

This seems odd, I know. Folks cannot run this. MCE::Simple will be released soon. I want to share the performance uptake having a limiter.

use MCE::Simple -strict, -signatures;
use MCE::Barrier;

sub task ($barrier, $number) {

    for (1 .. 400) {
        say "$_: $number";
        $barrier->wait;
    }
}

sub main () {

    STDOUT->autoflush(1);
    my $num_workers = 500;

    # create a barrier
    my $barrier = MCE::Barrier->new($num_workers);

    # create the worker processes
    spawn task($barrier, $_) for 1 .. $num_workers;

    # wait for all processes to finish
    $_->join for MCE::Simple->list; # same as sync;
}

main();
[download]

Results

Python3 and Perl complete in 4.1 seconds. That was with the limiter commented out. The time is 1.5 seconds with the limiter enabled.

sub new {
    ...
    # [10] limiter
    (!$is_winenv && !$tid) ? MCE::Semaphore->new(3,1) : (),
    ...
}

sub _wait { 
    ...
    $obj->[10]->down if $obj->[10];    # limiter
    my ($parties, $synced, $timed_out, $aborted) = $obj->_up($flag);
    $obj->[10]->up   if $obj->[10];
    ...
}
[download]

The other oddity is that I cannot spin 1,000 processes in Python. It complains about running out of file handles. Perl, no problem!