The benchmark is synthetic, so I can't be sure it will "feel" like it is working optimally, but it should be pretty representative as the clients are doing things (moving and placing blocks - generates cuboids and causes lots of updates) and also on a world which is simplistic, but does check all the boxes for world-gen overhead, storage cost, network interaction, and entity update processing.

Still, I would usually test with 10 clients on previous releases so pushing this one to 40 and seeing that it is still working pretty well is a good sign. I also maxed out the view distance for the headless clients so they are getting a lot of updates.

Software optimization is a never-ending task but this release made enough head-way that there are no obvious problems blocking further development, for now.