

I would put truenas on the NAS, also put a VM on truenas with 16-24G of RAM.
Create a kubernetes or docker swarm cluster with server 1 and the nas vm and just have everything as containers. This way you just have one resource pool, and the containers will be started wherever there are enough resources available. The containers will mount NFS shares from truenas which truenas will create automatically as ZFS datasets. ZFS supports snapshots.
Check out EFF cover your tracks: https://coveryourtracks.eff.org/
The results are very interesting. For me, the most unique thing about my browser was that I had two system languages, and so the accept-language header was very unique.
I now use vanadium (graphene OS), which simply sends made up values for a lot of headers, and so makes fingerprinting harder.
In general, you should try to be as “normal” as possible, use standard settings for everything, just accept English, etc…