It does sound good. Then again I thought using 11 different algorithms chained together sounded good....
At the end of the day if someone wants to build a device that can do it more efficiently than the average ETH miners PC then they will. Sure they might need RAM, they might need a copy of the blockchain, they might need some computational power....but it can be done. I'm sure it WOULD be done for ETH at the current market cap if they weren't planning to move away from PoW mining.
The "poison the well" thing is interesting. I'm curious as to how that would work and how it would be able to disadvantage a next-generation ASIC without affecting the average miners PC. It would need to be able to differentiate between the two. I am sceptical.