You can't reach the pure HW metrics due the VSAN SW stack and checksum calculation and .....
Your disk group design isn't so good. A lot of capacity Flash Disks.
Better choose a design with two or three DGs.
If you have 250K front end IOs you have mor than 500k Backend IOs.
Configure HCIBench to produce 10000 IOs and than see R/W latency....
I have in mind in the near future, split the discs into two disk groups, adding another cache disk.
I don´t understand well when you say:
"If you have 250K front end IOs you have mor than 500k Backend IOs."
I think that 250K iops (using 4 optane P4800X 375GB from caching) , is a poor performance, don´t you think?
These optane disk promisse a much better performance than standard SSD disk , but in this case a don´t get great results.