Above quote is trolling.
What I offered is a reality check. Apart from that there is no need for me to bloat or gloat. And for sure no need to proove anything. Take it or leave it.
Window 11 Enterprise (2x MadMax) = 82 PPD
Ubuntu 21.04 (2x MadMax) = 162 PPD
2x MadMax = 2 plots in ~18 min
1x MadMax = 1 plot in ~14 min
I noticed that the plotting time on 1xNVMe+RAMDisk is the same as on 4xNVMe RAID0+RAMDisk = 14min. Do you have any suggestions for swap size?
Question is: what type of ramdrive?
Mdadm is poor performance-wise. With latest changes the read speed of mdadm increased noticeably and scales well with threads, but write speed is still poor. So you are better off using single drives for Chia. Alternatively you use eraraid from raidix, which is a blast for conventional plots, but madmax cant handle eraraid and stops with fopen error:
https://github.com/madMAx43v3r/chia-plotter/issues/925
Until the dev fixes it, you are better off to use eraraid with classical GUI. eraraid + classical gui usually outperforms madmax plotter - slightly, but not much.
Old/wrong version - edit starts thereafter
You can do app 96 plots parallel with 2 threads each on a 3995WX. As long as your ASUS board holds up, you should exceed 200 PPD. Depending on final copy you need to start plots with a 15 minute window. This way you should get to 9-11 hours per plot.
Generally in official Chia you should be able to run 192 threads parallel on a 3995WX (docker, preferably rootless podman is your friend). And generally 200+ PPD should be possible. Problems here are:
- How many plots can your SSDs support each?
- RAM - memory requirements?
- Stage cascaded start
- Threads per plot
- Avoid blockade on destination drive
- …
There you need testing on how plot time and threads correlate for your system.
Thats the short version 
On swap size: you only need it if you are low on RAM and cant buy more. Once plotting goes to swap space you are doomed anyway.