If Bitcoin can change its PoW and retain market dominance, that means its value isn't intrinsically dependent on the mining infrastructure. If that is the case, then future PoW changes remain a permanent option, and the threat of using that option severely discourages future collusion. With collusion no longer a threat, ASIC resistance is no longer a pressing need.
In other words, a successful PoW change -- where success means the value of the network is maintained despite the change -- effectively solves the mining centralization problem because centralized or not, miners will know that any meaningful attack will be met with financial ruin. This becomes even more true if the mining infrastructure is reliant on ASICs and is therefore unable to switch to the new mining algorithm. ASICs are actually preferable if a PoW change is a constant and real option precisely because a PoW change effectively bricks those ASICs and severely damages miners financially. So a successful PoW change in itself would be sufficient, changing to a more ASIC resistant algorithm would probably be superfluous.
The problem we face today is that few people see a PoW change as a realistic option. Many are of the opinion that Bitcoin is too dependent on its mining infrastructure to divorce it without severe consequences. In many respects, the S2X fork will help clarify this matter, potentially even without a PoW change. This is because the gamble of S2X is that by starving the bitcoin network of hash power, and even potentially attacking the network to make it temporarily unusable, the market value of bitcoin will plummet. If they are wrong, then they have demonstrated that the value of bitcoin is largely independent of its mining infrastructure. Once that is understood, changing the PoW algo becomes a more realistic threat in the future.
Right now, one man in china with enough authority can order a 51 percent attack to do the following:
Reverse transactions that he sends while he's in control. This has the potential to double-spend transactions that previously had already been seen in the block chain.
Prevent some or all transactions from gaining any confirmations
Prevent some or all other miners from mining any valid blocks
As a handful of Chinese citizens operate the majority of hash power, it would be easy to perform a synchronized attack that stops other non Chinese mining pools from mining blocks, or a portion, enough to reaffirm majority at all times.
What good is having "a lot" of asic hash power if it's made by one Chinese company that could be nationalized or controlled? Which is not unreasonable given what we have seen happen to the Chinese exchanges under the guise of "KYC/AML" concerns, controlling the miners is also a KYC/AML concern for the PBOC.
Spare me with "there's other players, not just bitmain".. If you want a miner that is half the efficiency as a Antminer, sure you can buy from someone else. Mining is a race, the fastest player wins, the fastest manufacture wins. That is Bitmain by a long shot with no end in sight. Even if there's a new player what makes you think they'll leap frog Bitmain? Bitmain will make faster chips too. The reality is the ASIC centralization is a problem right now.