Actually, you have 2 single points of failure at the moment: the nfs server and the esxi host (and I'm not counting the single switch you have connecting them).
You should "future-proof" your system as best you can.
It is better to have several lower end servers in a cluster than a single "enterprise" server.

It also depends on the server. Most enterprise servers have dual power supplies, dual network connections...

I just thought of something, if I go the pxe route it means it's being done in software and not hardware (whether it's virtual or physical). This means the OS will see the nic that goes to the SAN. This is a HUGE security risk especially for VMs that are considered untrusted (ex: internet facing stuff, testing spyware/viruses etc). How would I mitigate that? NFS is not any more secure but right now it's only the VM hosts and VMs on main vlan that see it, not each VM.

And yeah my file server has redundant PSU. I have not gotten around to setting up network teaming yet but it's something I want to do. I could go all crazy and make everything redundant switch and all but I have to consider the best balance of cost, power usage and uptime as well as complexity. For a home setup I'll probably go as far as having multiple VM hosts and 1 file server.

I'm even contemplating just sticking with 1 VM host and 1 file server and not even worrying about finding a way to share the iSCSI. I probably wont be adding another VM host any time soon anyway and when I do I doubt I'll be able to match the hardware of my existing one as stuff changes all the time. So not like I'd be able to do live or even cold migrations or anything like that.

It does indeed, but an OS failure on that server will leave you dead in the water, no matter how many PSUs or NICs it has.
Also, it has only one motherboard.
Sure: it has a much lower probability of going down than commodity servers... but if it does, so will your whole system.

I have a client that, despite our recommendation, went with a single enterprise grade server (my company counseled 2 of them).
Although the server had 2 PSUs, they were connected to a single UPS...
Can you guess what happened one stormy night over the weekend?
If your answer was "the UPS fried" you won a cookie

That is a good point.
Yet one more reason to not expose the iscsi targets to the VMs.
I was only considering performance.
As I told you before, the network will always be your bottleneck, so I'd avoid using things like iscsi and nfs as much as possible on the VMs themselves (you kinda have to use it on the hosts if you want HA and migrations).

For a home test lab, you can have 2 hosts (don't do the file server / SAN) for about 250/300€. That is the price of a decent smartphone nowadays...
Sure, it will be commodity hardware that will run 2 or 3 VMs maximum per host... but still, it will be quite enough for testing things out before you deploy the system on production servers.

The advantage of exposing iscsi and NFS to the VM is that when the VM moves (and you don't get the confusing problems like the OP had)... so moves the IP address used for both, in the same operation.

With only the host have access, then the IP addresses don't move. AND you get the additional overhead imposed by the host.