I would argue (and personally prefer) the opposite. The OOM killer reaping your application is easy to diagnose (it's logged) and it is obvious what you need to

I would argue (and personally prefer) the opposite. The OOM killer reaping your application is easy to diagnose (it's logged) and it is obvious what you need to do to fix it. Random slowdowns due to it hitting swap could be very hard to debug if they cause issues.