100% this. Sometimes it's not even the filer itself. `hard` NFS mounts on clients in combination with network issues have led to downtimes where I work. Soft mounts can be a solution for read only workloads that have other means of fault tolerance in front of them, but it's not a panacea.
I haven’t seen these problems at much larger scales than are being discussed here. I’ve heard of people buying crappy nfs filers or trying to use the Linux server in prod (it doesn’t support HA!), but I’ve also heard of people losing data when they install a key value store or consensus protocol on < 3 machines.
The only counterexample involved a buggy RHEL-backported NFS client that liked to deadlock, and that couldn’t be upgraded for… reasons.
Client bugs that force a single machine/process restart can happen with any network protocol.
