some kinda synchronized processing

Er...threads have enough problems with non-deterministic behavior, you might want to refine that "requirement" a bit ?

And I spotted a rather gruesome wart in the midst of your example: invoking

system("date")

from inside a thread is not exactly best practice (you do realize system() turns into a fork()?). Consider using

print "test\n", scalar localtime(), "\n";
[download]

instead.

Of course, making that edit changes the whole behavior of the threads: since its much faster, many threads will complete before other threads have even been started...so its not exactly simultaneous.

As BrowserUk mentioned, look into lock() and cond_wait().