First cache pool is the write tolerant Optane 750 GB with ~30 DWPD or better - IDK the figures off the top of my head.. Second pool for overflow is a much less write tolerant Nytro 1.6 TB with 3 DWPD. Then it hits the main storage - 100ish TB of spinning rust.
By doing this I can use the Optane to cache all the VM writes and a Chia plotter or two, with the Nytro catching any overflow as well as the written out plots
Unraid allows you to cache writes for shares, writing them out with the mover daily or as configured, as well as setting preferred locations for certain types of media, such as the VMs and (unraid) appdata, sometimes moving them to that preferred location depending upon configuration, or simply caching the writes.
It's certainly not optimized for the fastest possible use, but it does allow me to flatten out my write to disk curve by a lot on each form of media, and speed up the work I do in general - at least in theory. Let's hope it pans out because I refuse to destroy regular SSDs.
I rarely revise content written to disk, just the VMs, so this could work very well for me optimizing things with the limited resources I have.