Update - Disconnects and Latency Issues

Let me put it to you this way

You have two options:
1) Every time an object comes into existence, you scan every single other object in existence, across hundreds of billions of them, to detect ID collision.
or
2) You don't commit saves until all steps of the save process have completed successfully, and if any point of it fails, you rollback the entire operation.

Which of those is more painful to do hundreds of thousands of times a minute?
Is it the one where you parse terabytes of data over and over and over? Or the one where you don't do anything but what you would normally do, with the added caveat that you do absolutely nothing at all if anything fishy occurred?

Question: using 2), wouldn't the server need to keep track of all such transactions until they clear? and then save them one by one? You mentioned saving on quit, but it would seem risky and a potential source for data loss, especially if there is a certain accumulation of transactions. You don't have to explain, but I'm quite interested, so feel free

PS: Sorry for sidetracking here.