I just absolutely dont get who would want this instead of RAM.
Everyone who understood that normal RAM is a really bad idea, because Machine failure / Power failure will lead to data loss.
Redundancy? Put the data to two severs simultaneously.
ZFS is (typically) not a Distributed filesystem. ZFS is not Ceph. And even if it worked with ZFS, you're still gonna have a bad day in case of a power failure without UPS (both servers failing at the same time)
Adding another server also introduces the requirement for a quorum, which adds a lot of complexity.
Plus you double the power consumption by adding a second server...
Put it to nvme. Those classic mlc/slc nvmes are only few times slower than RAM.
"only a few times"... Dude, are you even trying?
The difference in IOPS between RAM and SSDs is MASSIVE.
There's a good reason why the only SSD (sensible) people use as SLOG are Intel Optane SSDs (like the 900p). With almost every other SSD as SLOG, your storage is gonna be dead slow.
What you want as SLOG is basically
1. NVDIMM-N (e.g. Micron NVDIMM-N, Radian RMS 200, etc...)
2. NVDIMM-P (e.g. Intel Optane DCPMM)
3. Optane SSDs (e.g. 900p)
(ordered after decreasing performance. Normal SSDs include because that's a terrible idea)