Maybe for maintainence purpose or schema change. For example as you mentioned above, if using several tablespaces located on different hard drives. If one hard drive is damaged, the database on that hard drive will not be available. How about other databases managed by the same database server instance?
interesting questions...
If I were you, I would just verify this by creating separate partition, putting a tablespace there, creating a database in this tablespace, pgbench'ing this database continuosly, and running umount --force on the partition.
I need to shut down all the databases to do mainatainence, right?