The lack of synchronization in the script enables the interpreter to switch threads in the middle of the computation of the f function. This race condition sometimes makes the script to report the wrong result.

You can solve this issue by applying a lock:

import threading

def concurrent_f(x):
    global result
    with lock:
    	result = f(result)

lock = threading.Lock()

Or by increasing the switch interval at the beginning of the script:

import sys
sys.setswitchinterval(1) # The default is 0.005 (5 ms)

This is not ideal solution but it doesn't modify the original concurrent_f function.