Obviously, this refers to how automated ‘decisions’ are made by the software, and does not mean that hash power gives you weight in social decisions in the real world. More importantly, there are two levels of decision making that people tend to conflate.

Hash power is relevant to 'regular' forks, as the longest chain that follows the rules will be followed. However hash power is not relevant for deciding what those rules are. That is done by full nodes, whether miners or not, and doesn’t depend on costly work or anything else. Nodes that don’t agree on the rules just stop talking to each other, and the network splits.

Then, the final decision about which network and chain trades as “BTC” is made by ordinary real world consensus, probably mainly among exchanges, wallets, and users in general.

Edit: Changed 'soft fork' to 'regular fork' because a soft fork is a protocol change. POW chain-following doesn't even rise to that level.

I actually have a lot of sympathy with this point of view. But there's a counterargument: if one fork has much greater hash power than another, then the minority is vulnerable to 51% attack from a fraction of the majority. So it's not like we can just ignore mining power, unless we make a fork that changes the hash function.

For Bitcoin, the difficulty adjustment would also need to be sped up, to keep the minority chain from being very slow for a while.

This is all true. The way I think about it is this would be just like any other 51% attack, except that the motivation is specifically to compete by killing the former network. But game theoretically, it's still an attack, not a valid way to achieve consensus. So, as you say, the correct response is for the original network to hard fork.

To be clear, I'm not saying 2X is itself an attack (IMO, awfully close to crossing the line, though). I'm assuming that in your scenario, some mining faction is spending part of its compute on running 2X, and part on explicitly 51% attacking the original network.