A timestamp that is too far into the future (compared to arrival time) is provably not an honest (correctly-functioning) node and needs to be rejected more strongly, possibly permanently unless the receiving node discovers his clock was wrong. Unreasonably long in the past is assuming there’s not large propagation delay so it must go into a timeout in case there was a real delay. PoW eventually discover which partition has the highest hashrate. If applying your idea in the case of old timestamps, then a new block with a good timestamp that comes in will be working on top of the bad timestamp just like you unless there is a real network delay. This allows the normal selfish mining if there’s not a real network delay because the miner could be withholding the block to get a “head start” on it. And if there’s a real partition, it’s not clear to me that you should switch.