Indeed, You may use values off, 0 (zero), false, or no to turn off the synchronous_commit. As the name indicates, the commit acknowledgment can come before flushing the records to disk. This is generally called as an asynchronous commit. If the PostgreSQL instance crashes, the last few asynchronous commits might be lost.

Regarding your global architecture, Multi-AZ offers a sync replication between the instances which may slightly degrade the performance in OLTP system.

If your RPO is 30 minutes, I would suggest to use read replica instead of multi-az while keeping synchronous commit at the primary database. Read replica is asynchronously replicating the data to the secondary DB. It may be enough to fix the performance problem.