Because the powers that the miners have are a lot less than the powers that the banks have. Miners, even at 99% hashrate, are unable to change the rules of the network. The most they can do is censor transactions and spend their own money multiple times. But even a miner spending their own money multiple times costs money - you have to redo blocks (costing tens of thousands each).

Banks on the other hand can print money, change the interest rate, steal, block people from the system for arbitrary reasons.

Most people don't realize that a 51% attack is not a 'we get to do anything we want now' attack. Miners still need to follow the fundamental rules of the network, and they also need to blow huge amounts of cash on electricity to maintain the attack. Today that would cost millions of dollars -per day-. That money needs to come from somewhere, and generally that comes from selling the coins that the miners get paid. If they make the system useless, the coins they get won't have enough value to pay the bills.

So there are a lot of dimensions that mean a cartel of miners could reasonably be construed as an acceptable alternative to banks.

Are you referring to Bitcoin?

In that case, you're talking about mining pools, which are different from miners. Individual miners participate in different mining pools, and can choose to use different pools at any time. Pool operators don't have that much control over the mining power, as miners will readily use a different pool if it's seen as a better alternative.

Historically, any time that a particular pool gets large enough to be a concern, miners switch off from using it as the risk to the network easily outweighs the small marginal benefit they may get from using a popular pool.

Miner centralization is certainly one of the primary risks in the Bitcoin network, but it's also one that's easy to overstate. Even with fairly extreme miner centralization, it's hard to come up with a good way to abuse that power without the rest of the network participants simply forking away from you.

There's good reason to think that the current level of geographic centralization in China is likely a maximum. The rapid development of ASIC hardware has kept investment timelines short as old hardware is rapidly obsolete. As hard limits in IC design are reached, this development has slowed, and may result in more competitive opportunities in other areas.

You don't have to trust the miners if they don't follow the rules their blocks are automatically discarded. The interesting thing with Bitcoin is you don't have to trust anyone else. It's also a push model vs the pull model of traditional banking.

You have to trust those miners in aggregate -- they have the power to change the rules, effectively, since the vast majority of network users will follow the majority of hashing power.

Similarly, if any one bank doesn't follow the rules, they get kicked out of the financial network, to similar effect. You don't have to trust a bank, you trust the banking (and legal) system.

If these malicious miners you speak of "change the rules", it creates a fork in the blockchain. Even if they have a majority hash rate, if the economy does not support their decision, the market value of that forked chain will be low while the unchanged fork will keep the market. A group of malicious miners can do very little to destroy bitcoin.

Depends on the sort of change, doesn't it?

Isn't that the soft-fork / hard-fork distinction?

It is true that miners in general cannot use their hashpower to force a hard-fork, but they can, I think, cause a soft-fork.

And, soft-forks can do a lot of stuff.

In what way does the same logic not apply to banks?

Banks can position themselves to benefit from a depreciation of any given currency, miners are tied to the success of the one they've chosen.

Can a miner not distribute their effort and holdings?

Miner hardware is tied to the specific proof-of-work algorithm used in the cryptocurrency they mine. They make extensive capital investments that are, at least in the case of Bitcoin, useless for any other purpose.