ZHA extremely slow and unreliable with ConBeeII stick

That would be a much more stable solution because not tunnel Zigbee serial protocols over a network. Such a solution cannot be compared with serial connection to a “network-attached” Zigbee Coordinator. If you link a second Home Assistant instance with a “direct-attached” Zigbee Coordinator USB adapter then that would convert the serial protocol to MQTT or other protocol meant for network communication.

There is more discussion about it here → https://github.com/home-assistant/architecture/issues/246

There should not be any serious issues in practice if have a wired Ethernet “network-attached” Zigbee Coordinator connection over a local LAN. Many people in this community use wired Ethernet connected network-attached Zigbee Coordinator adapters like Tube’s Zigbee Gateway and ZigStar LAN Gateway without any problems at all. You can read many reports from their users in the HA community here:

https://community.home-assistant.io/t/zigstar-zigbee-coordinators-and-routers/338586

https://community.home-assistant.io/t/tubes-zb-coordinators-and-routers-was-zigbee-router-on-steroids/280896/

Also, note that based on the fact that there are fewer reports about it in the communities, Texas Instruments ZNP serial protocol is generally believed to have slightly more robust than the Silicon Labs EZSP (EmberZNet Serial Protocol) in the way that Silicon Labs based “WiFi-attached” Zigbee Coordinator connection is infamously known to be notoriously unstable when connected over a Wi-Fi network that is not extremely stable. Though to be fair there are today a lot more Silicon Labs WiFi-attached Zigbee Coordinator products on the market that has been hacked for use with ZHA and Zigbee2MQTT than there are Texas Instruments based WiFi-attached Zigbee Coordinator products).

I think that in practice the only common scenario that causes issues for even Texas Instruments based wired Ethernet “network-attached” Zigbee Coordinators is when the LAN/switch does down for whatever reason (like a locally blown fuse) and your ZHA/Z2M instance is still up, then it would be like pulling out the adapter during operation. That would be to the effect same as in a scenario where you would pull out a Zigbee Coordinator USB adapter without stopping ZHA/Z2M. In that case and the wired Ethernet “network-attached” Zigbee Coordinator comes back then you might need to restart your ZHA/Z2M instance afterwards if the serial protocol the Zigbee stack uses is not robust enough to recover.

To workaround that issue you could as a suggestion make sure that both your computer running ZHA/Z2M and the wired Ethernet “network-attached” Zigbee Coordinator are both powered by the same PoE (Power over Ethernet) switch as that way if one dies due to power loss then other also dies too.