OP is really just asking would this be possible as asking that question is usually the best way to think of ways to combat an attack. Personally, I don't think this would work in the least as you'd really have to collude with a pretty good amount of miners in order to make this is a reality,

A good majority of the people that run these huge mining farms and pools would probably have some amount of ethics and put these above any amount of profit, not all though. Greed could take over and this could work but I would hope that this would never occur.

For this to work, you would have to have close to 100% of the miners colluding with you. If not...

The blocks that are not mined by the colluding miners would be full of high fee paying transactions that are all legitimate transactions in the eyes of the non-colluding miner. While the blocks mined by colluding miners are partially full of legitimate high fee paying transactions and partially full of transactions crafted by the colluding pools to increase the overall transaction fees.

Therefor, non-colluding miners would be earning more per block as all of the fees in their blocks are actual income. The colluding miners would be earning less per block because some of the fees in their blocks are just their own money being returned to them.

Eventually the non-colluding miners would be able to purchase more hardware with their greater profit and slowly erode the colluding miners percentage of the network.

If they mine only their own tx's to cancel out the cost, they would eventually run out of money. Mining is expensive and they have to make a income to make their mining operation sustainable. An attack like this will only decrease their profits and give other miners the opportunity to surpass them.

A lot of the profits are spend on expanding the operation to accommodate for difficulty increases, so they will lose a lot of money, if they spend it on worthless attacks like this. ^hmmmmmmm^

This is enough for making profit. No need to play child games with fees.

A 51% attack is a once-off basically. If you do it successfully, the users will simply swap to a different mining algorithm which will render all of the attacker's expensive hardware worthless, so you had better make bank.

Looking for less disruptive, sustainable ways to earn extra is therefore more sensible.

That said, this particular "attack" is pointless for the reasons I mentioned above.

You assumption that mining pool has its own hardware.
The 51-attack would cost nothing for the pool admins if they do not have their own hashpower.

considering the miners are making about 2 or 3 times more profit compared to before you can make that assumption that a miner or groups of them are doing this.

but some things don't match in my opinion.

if it was miners they wouldn't have spammed the hell out of it (it is 90K to 100K transactions today) while only 10K is more than enough to cause a fee war and increase their revenue and keep the cost to a minimum.

also the time this nonsense started and times when it goes up matches mostly with debate over block size and that can not be a coincidence.

and also as a miner you must be really super greedy to want more fees because the block reward and the price is giving them more than enough to cover their costs and give them a huge profit on top.

Again, this "attack" benefits the pools/miners who aren't involved far, far more than those "mining their own fees".

Non-colluding miners get blocks full of real, fee paying transactions. Wow, they love this "attack".

Colluding miners get blocks composed of some real, fee paying transactions and some home brewed transactions with fees going back to themselves. Meaning, they get less real income from real, fee paying transactions.

Do this "attack" long enough and the other miners will simply render your hash rate irrelevant as they have more profit to buy more hardware.

I suggest you follow your own advice and "think about it". I won't even mention what happens when one of the colluding miner's blocks is orphaned and their "fee to themselves" transactions are known to the entire network.

how can users simply swap to a different algorithm when it require an hard fork, and there will be never a consensu for this?

if the 51% of nthe net decide for a 51% and the remaining not, i doubt those 49% are going to agree to lose all their asic sha256 for a new algorithm