Several years ago, I had a customer with to copies of a database – one “current”, and one where they ran another copy of the application to try to catch up on data entry for past years.

Of course, once they finished the “past years” data entry, they wanted to merge the two databases. Of course, the two databases both used sequences to generate primary keys, and both started the sequences at the same starting values.

Also, of course, I had to advise the customer that it wasn’t worth in terms of cost to do the merge, or the danger to the integrity of the data.

Ever since, I’ve used UUIDs whenever I can.
I’ve never seen a performance hit for doing this, but then, I’ve never worked on a VLDB