Very interesting insight! However I'm not fully behind some of the points.
Saying that fixing a bad hire is easy is a very USA-centric view of the world. In some European countries it would be very hard to get rid of such people, so the cost is hire.
I think the S3 analogy is not apt either. Sure, the probability of an error for each individual object is negligible, but if you store billions or trillions of objects, it becomes certain.
A better tech analogy is this: imagine a service that creates a bunch of resources every time you create an entry. When you delete the entry, you want to get rid of each resource. Now you have to defensively program around the myriads of errors and possible inconsistencies that can arise from this. And of course everything can just go wrong anyway.
Instead you could do nothing, or an optimistic delete. Then have a cron job that periodically cleans up orphaned resources. This is a much simpler and resilient approach.
The right answer to this question can probably be semi-formalized by account for the volume of transactions (interviews), errors (cheating candidates), cost of error (bad hire), price of increased complexity (more cumbersome interview experience for all), and how effective it is at preventing errors (are we getting all the cheaters?).
EDIT: On an extra note, this reminds me of those anti-piracy measures that don't stop piracy but are a royal pain in the ass for legit buyers.