As an afterthought, when I first tested your original script, I tried to redirect the output to a text file with " script > textfile". I did lose some lines that way on repeated attempts, due to the script exiting before all output was printed. Maybe there is some sort of timing problem, between the ParallelForkManager, the IPC, and the main script. Have you tried putting delays in there, to let one part of the script catch-up with the other part? In my experience with other scripts, where one person sees an error, and another dosn't, it often involves timing differences related to the "speeds" of the different machines. I've even seem cases where scripts will fail when the machine is under high load, but run OK otherwise. Usually, slowing the scripts down, helps. Maybe the IPC through shared memory is "too fast" for the big modules to handle in perfect synchronization?