That would create a fork that nobody would use (because bitcoin holders are incentivised by the halving mechanism).
Except the "small group controlling 90%" of mining power that introduced that change (in the hypothetical example) would use it. And their chain would grow longer and be the canonical one (though that, of course, is also just a convention encoded in the code). And they could still spend some 20% or so of their hash power to entirely mess up the one true chain, if they so desired.
You would need to control the majority (of the important) nodes to be able to pull this off. If the nodes are incentivised to keep the limit, they will.
The history of the bitcoin blockchain can't be changed unless a miner can contribute the same amount of work it took to generate the section of blockchain they'd wish to change (so from a historical start block to the latest block). Even then, the most they can do is undo payments and double spend coins. This would probably not be economically viable since Bitcoin clients would probably start to blacklist expenditures from the miners addresses if suspicions of a double spend attack arose. Because wallets and nodes can tell when there's forks, people would also pick up on double spends which would disqualify any and all money attempted for the double spend. Overall it would probably be cheaper to just spend coins fairly.
Also, as an aside, the protocol-valid bitcoin blockchain with the most work gets chosen, not the longest* blockchain