It appears to me, for what I'd read in this thread, time-dependant unique id's are only possible in a perfect world, but what about the real one?
For example, if the workstations should periodically synchronize their clock with the server, or some external reference (sometimes necessary) then their clock can go backwards then the possibility of creating a duplicate key is not zero and will be proportional to the adjusted time (for example if the adjust was only .1 second the collision will be less probable than a 10 minute adjustment).
If for system time you are referring to the time on the server, but your server is using some external time reference, you end up with the same scenario.
Then, don't you need some verification process to avoid this?
"The five senses obstruct or deform the apprehension of reality."
Jorge L. Borges?
"Premature optimization is the root of all evil in programming."
Donald Knuth, repeating C. A. R. Hoare
"To die for a religion is easier than to live it absolutely"
Jorge L. Borges