How do you effectively determine if the massive cost savings of Preemptible VMs justify the risk of your workload being terminated at any moment? Furthermore, these instances utilize Google's excess capacity to offer discounts of up to 91% compared to standard rates. Why is architecting for fault tolerance and using tools like Managed Instance Groups the most critical priority when deploying these short-lived resources?