Seems that when running on larger datasets- join group stability memory load is quite high, and so when running in parallel can overwhelm the computer