It's the default setting that didn't cope well. The final setting seemed to get pretty decent performance. And if you only have one knob then the rabbit hole is pretty shallow.
The question not answered (as it wasn't asked) is how many workload profiles can be catered for? But as a sample of one, this appears successful.
The question though is does the setting that worked for this particular micro-benchmark actually work well for more generic workloads that have similar performance issues.
The question not answered (as it wasn't asked) is how many workload profiles can be catered for? But as a sample of one, this appears successful.