Hmmm, that 3.5GB/s sound low. From rough memory of doing initial storage benchmarking of our "new" Hetzner dedicated boxes a few months ago (AX51-NVMe, https://www.hetzner.com/dedicated-rootserver/ax51-nvme), they were giving about 10GB/s with mirrored NVMe drives.

Just logged into one of those boxes now, and it's running 2x 1TB Samsung PM9A1 drives (https://semiconductor.samsung.com/ssd/pc-ssd/pm9a1/), compression is on (lz4), and dedup is off.

(Didn't do any real tuning at the time, as these specs were already far in excess of what's needed for these servers.)

Would enabling lz4 compression be useful for your use case?