There seems to be some confusion around how ORDS works with it's connection pooling yet running the REST call as the specified schema.

The connection pool

Consider a 50 PDB env and concurrent users per PDB running some REST stuff.  Using a connection pool per PDB would be 50 connection pools.  Then if a JET app ( or any HTML5/JS/.. ) is making REST calls Chrome will do this with 6 concurrent calls.  This makes it so that as simple as 5 concurrent users * 6 threads in Chrome * 50 PDBs would reach 1500 connections.

For this reason, in a CDB install the connection pool is established as a common user in the CDB names ORDS_PUBLIC_USER.  This allows for common connection pool that can be shared over all 50 PDBs.  As requested finish, the connection can be recycled back for any PDB to service the next request.

What about Auditing 

This is where the confusion begins.  If it's a common connection pool, it's un-auditable. FALSE

I made a trivial REST api that does nothing more than a select on EMP

Then I turned on Auditing of this table with the following

ON klrice.emp

Running the REST call and checking the DB level audit results in exactly what is expected.  There's no notion of the common ORDS_PUBLIC_USER anywhere.  The user KLRICE accessed this table making all normal DB auditing work well normal.

Then why all the FUD?

I think the confusion is that APEX using a different mechanism than ORDS for executing the user's sql with the parse_as options.  When run like this the object audit log shows that the common APEX_PUBLIC_USER is the user that accessed the object and the parse_as user of KLRICE in this case is not present.  However, APEX does populate the CLIENT ID so the audit trail has the runtime user that accessed the object.