When you do a fast stop, all users have their transactions aborted and all connections are disconnected. This is not very polite to users, but it still treats the server and its data with care, which is good.
PostgreSQL is similar to other database systems in that it does do a shutdown checkpoint before it closes. This means that the startup that follows will be quick and clean. The more work the checkpoint has to do, the longer it will take to shut down.
One difference between PostgreSQL and some other RDBMSes such as Oracle, DB2, or SQL Server is that the transaction rollback is very fast. On those other systems, if you shut down the server in a mode that rolls back transactions, it can cause the shutdown to take a while, possibly a very long time. This difference is for internal reasons, and isn't in any way unsafe. Debian and Ubuntu's pg_ctlcluster supports the --force option, which is rather nice because it first attempts a fast shutdown, and if that fails, it performs an immediate shutdown. After that, it kills the postmaster.