Same - I had a reply but I didn't like how it sounded so I'm still drafting, lol
the entire relay network can be the same .. zero conf won't be reliable.
The network cannot be the same because there is latency between nodes, thanks to the awesome principle of the speed of light, processing requirements, memory latency, utxo db lookup speed... and so on. Therefore zeroconf cannot be 100% reliable, because even if you see everything, you cannot know what is included in a miner's template that got distributed to all these ASICs, or maybe you only see it after a block was already found... So even in a tx replacement fight, you need luck. It's a race.
Note that I don't really want to be discussing stuff like this in all its glorious larp-cases in the open if you don't mind - the defense is always aggressive replacement / rebroadcasting and mempool monitoring. And being aware of what you're using your node for. ↩
Footnotes