If it's running inside a hypervisor, does the hypervisor not present switching capability to the guests, obviating any need for the guest itself to do any switching? I know VirtualBox, VMware Workstation and VMware vSphere do. I don't - yet - know about Proxmox's capabilities.
why this attitude ?
i am using in everyday job tp-link jetstream (rack mount) switches, for years
and no, no crap, everything is working as expected, snmp,lldp,ddm,sfp,vlans ...
or Eltex 23xx series, they are working for years with zillion bytes of multicast traffic. And no, nothing fancy, very minimalistic WEB, and all options what good L2+ need to have
so it was very blunt statement
no, really there is no switch menu
there are trunk ports & bridge vlan filtering in new OWRT releases
so, simply add ethernet ports to Proxmox bridge, and attach OWRT to this same bridge, then use vlan filtering to get "switch" feeling
If OpenWrt has direct access to the card it is always going to look like four CPU ports because that is what it is. A multi-port card has no "data plane" hardware path for data to flow directly between the ports like an actual switch has. A switch would be simulated by building kernel bridges between the port drivers. This is still quite fast.
I would recommend against such a configuration. A hypervisor shouldn't be dependent on a guest for networking; doing so could lead to some rather interesting "chicken and egg" scenarios. Rather, the guest should be dependent on the hypervisor.
What you propose can be achieved, with care, but it's not an optimal approach.
Whichever is better-supported by the hypervisor and is within your desired budget.
Personally, given the difference in attitudes towards open-source displayed by the two companies in question, I would veer towards Intel. Other people may have alternate views.