When I chose XORShiftRandom for my MCTSs (over SplittableRandom…) it was based only on internet searchs and submits in the arena.
I did several benchmarks using Java Microbenchmark Harness :
| Benchmark | Mode | Cnt | Score | Error | Units |
|---|---|---|---|---|---|
| SplittableRandom | avgt | 25 | 246045,663 | ±27970,447 | ns/op |
| XORShiftRandom | avgt | 25 | 357386,862 | ±46990,309 | ns/op |
SplittableRandom performs better (but not 4 times better).
There a Stack Oveflow’s post where @LoganWlv give a link to benchmark results and an explanation.
So yes SplittableRandom seems to be a better choice.