Well, for starters, -r 15 would help, probably quite a bit. Hopefully you’re firing off plotting from a power shell prompt, as it is cleaner and easier to do.
Set CPU affinity to the chia_plot instance(s) to 15 threads, as MM will try to take them all even though you spec lower. You could try K 2, might not help but worth trying. Also, try two instances (same parameters).
If you do do two instances (two PS prompt windows), starting a second instance after the 1st instance finished P1 will get you ~10% faster overall average plotting time on average. Guessing you’re making k32 plots, so your times should go down from what you got before.
Last, format your nvme with 64k or 128k sector size, not the default 4k.
Here is an example ssd read/write vs block size test. Notice that xfer rate plateaus at greater than ~64kb and higher block sizes. Smaller sizes have higher overhead, hence less data. To maximize write speed, but also minimize wasting space with too large block size I choose 64k or 128k.