I'm not sure you can achieve good event sourcing performance using a regular database engine. Better to view it like writing logs.

If you really must expose an SQL API, perhaps you could read the journal on another thread or process and then make changes to the db based on the incoming "diffs" that the threads determines from the journal?