For me, the key idea of this paper was that relational model & SQL are successful because of the key feature: 'compact and accessible form' of a query.
Hmm, I found it quite debatable. IMO, they are too bonded to the opinion that everything will remain almost the same in database UI and use cases. Looking around at how eagerly analytics employ AI-generated queries, I wonder if they have any arguments to support this standpoint.
The big problem of serials is their monotonic grow. Each time adding new record you go beyond statistics boundaries and challenge PostgreSQL to guess..
The term ‘slow’ is too relational and not strong. I guess, we should look up for queries, which can be potentially faster - see into estimation errors or number of data pages involved into the query.