My fuzzy take on a concern here could be: storage costs and memory usage. My impression is that it is written/copied to actual rows as xmin (https://www.postgre

My fuzzy take on a concern here could be: storage costs and memory usage. My impression is that it is written/copied to actual rows as xmin (https://www.postgresql.org/docs/current/ddl-system-columns.html https://www.postgresql.org/docs/current/ddl-system-columns.h...). So loading X rows from disk into memory is loading X transaction IDs into memory. That said, I'm not sure the concern holds water. At 2bil rows we're talking 7.4GiB of disk vs 15GiB. Depending on what % of your database you want to keep in memory, maybe it goes ok.