parallel computing of prime numbers

[meh <mehlng@yahoo_nospam.com>]

I'd like to compute parallely primes with variable amount of tasks. I 
made a task with an entry to compute a single number's primeness and 
then looped through all tasks and assigned each task a new number to 
work on.
My problem is that the assigning command waits until the prime 
calculation ends, I want it to assign the task a number to work on and 
then CONTINUE IMMEDIATELY assigning other numbers to other tasks.
I thought of making an array of semaphore and input variables, and then 
to use them to assign numbers to tasks (I'll intialize each task with 
the relevant semaphore), but it seems very clumsy way to do things.
Pray, how can I do that in a more elegant way?