Hi, I come late to the party but let me chime in.
If you really need performance and support for a heavy production environment, go the TrueNAS Scale route as @tangles suggests, but don't even bother with a cMP. There are many used Dell or Gigabyte servers you can buy that will work much better… but if you are doing it as home lab project or for a small workgroup/company, it is completely feasible to use cMPs.
My most beloved project is a cMP 5,1 12core with 96GBs or RAM with OpenCore Legacy Patcher to run Monterey, a Sonnet McFiver card for 10GB ethernet and two Samsung 990 Pro NVMe SSDs (I now, they are overkill, but the client always has what the client wants

) and an ATTO ExpressSAS PCIe HBA card connected to an external JBOD with 24x4TB HDDs configured with OpenZFS on OS X (4 RAIDZ2 vdevs), with L2ARC on the mirrored NVMe SSDs. It is working so well for a team of ten people that we are looking to upgrade the machine to 128GBs of RAM (the maximum) and try deduplication… I know, everybody tells me not to do it, but as I am going to have a duplicated setup like that one, I want to try the limits and even configure a metadata vdev and slog vdev on a 4 NVMe PCIe card (non bifurcation model) and see if it holds. Everything backs up to another two servers via send and receive and it is incredible to see backups (after the first one that we do locally and then take the backup servers to their final destination) happen in a matter of minutes across town.
At this point this project is just for the sake of trying, because if you take into account the price of the PCIe cards, NVMe SSDs, etc… you better go an buy a modern Intel/AMD rack mounted server with everything enclosed (right now we are using decommissioned SAS 6Gb JBOD boxes, so that part is really cheap). If the client lets me I would like to post some videos in Youtube explaining the project and showing it step by step.
If we finally decide to change solution then the magic of ZFS comes into play again: take the SAS JBOD boxes, connect them to a TNS, Linux or BSD server, import the pools and keep going on like nothing happened.
There are some 2019 MPs appearing in the used market, but they are still 1500€ for basic models. If they come down to 1000€ the client has allowed me to buy a couple of them and see if the performance justifies changing all the cMPs. I will make my best to show the improvement
Once those cMPs die because they brake, or probably because there will be a solution that doesn't chew so much energy and it is not justifiable the power consumption versus the performance, we plan to use the cMP cases to create some benches to sit on, so we can see them everyday and still admire their great design and usefulness over such a long period of time. Long life to cMPs