Yes indeed, let's do the math...
Suppose that your machine has a 1 millisecond timer-quantum, which means that you get 1,000 of these puppies per second.
And you have 2,000 active threads.
Worst case, every one of them gets to execute for 1/1,000 of one second every two seconds.
Read my lips: you
cannot create a "real world" server which depends upon "creating a thread for every request that it receives!" If you take this "flaming-arrow approach" in a real-world situation, you wind up with a bonfire.