> This is true for both a hard and soft fork. Longest chain wins, so per definition you "only" need miner majority, no matter what change you want to make.

Not quite.  With a hard fork, a minority of (say) 20% of the miners can continue mining their own branch.  Even though the branch with the majority of the miners has more proof-of-work and is growing faster, the minority miners are not required to accept it; because, to them, that branch is invalid.  (This is the technical detail that makes soft forks easier to deploy than hard forks.)

It is true that [there is no technical difference between soft and hard forks after the fork happens](https://www.reddit.com/r/bitcoin_uncensored/comments/3hj2sl/soft_fork_hard_fork/).  In either case, after the change is activated there are two communities of miners fighting to define the blockchain: R with the more restrictive rules, P other with more permissive ones.  If R ever gets a majority of the miners, it will win because the P branch will be obliged to use their branch, so it will lose any block that it mines.  If P gets a majority, on the other hand, the R miners can continue mining their branch and ignore the P branch.

So that is the operations difference between the two kinds of changes: with a soft fork, if the change ever gets activated, the new R set will start out with a super majority of the miners (unless they lied about their intentions), and the old P branch will die naturally, maybe after some large reorgs.  With a hard fork, the new P set starts out with a  supermajority, but that, by itself, will not kill the R branch.  To prevent a coin split, the P miners would have to actively sabotage the R branch, and/or convince enough services to switch to P so that the R branch will lose its value.

> exchanges and merchants have no reason to change software and accept them

Some of the largest miners in the world are connected to the largest exchanges in the world (Huobi = F2Pool, BTC-China = BTCC pool; OKCoin is connected with a pool too, but I don't know which one.)  So those exchanges, at least, will switch to whatever the majority of the miners decide.  And a majority of the miners can afford to sabotage a minority chain, so that its miners starve and its clients are unable to use it, if it needs to.

Anyway, the majority of the miners will only accept a change (no matter who proposes it) if they believe that the end result will not harm their bottom line.  This evaluation includes, of course, estimating of the price change that it will cause.  This depends on how users and holders will view the change; and this can be influenced by suitable PR, endorsement by hired gurus, etc.. 

After all, if someone can believe in bitcoin, he can be made to believe anything ;-)

Note that even if Satoshi hates the change and decides to dump his million 0.001 USD/BTC coins, buyers who do not mind the change will pay much more than that for his coins.  So, if the old holders dump, they will only be replaced by new holders who do not mind the change; and the price will be then determined by speculation among these new holders.

> because the old chain keeps working (thanks to the soft fork).

No, in a soft fork the old **chain** will die (repeatedly, after a few blocks) because of the MoW rule, but the old **clients** will continue to work in part and will not notice the change, except for unexplained rejection of transactions and incoming coins that do not seem to arrive.

In a hard fork both **chains** continue working separately in principle, and the old clients will see only the old chain; which will grow much more slowly, but without major reorgs -- until the client upgrades, when suddenly the entire old branch will be replaced *for him* by the new one.  

That is why, in a hard fork, all old clients **must** be alerted and given time to upgrade, before the change is enabled; whereas, in a soft fork, the clients can be casually advised to upgrade only after the change has been permanently put into effect.  And that is why the Core devs like soft forks and hate hard forks...