But I'm not sure if the fee of the replaced transactions that a node relayed wont be removed from its "bucket" and if those will still be counted to its fee estimation after the last replacement is confirmed though.
This is the final replacement that got finally confirmed:
https://mempool.space/tx/b5fe2c4891cba365c0863b8645a7e7a4c7e7191804508f4549f870be6012102fI've counted more than 30 replacements of the original tx that were done within 40 seconds intervals on average. This kind of txs have only one input and only one output. This is why they are cheap since their size is below ~190 vB. There are many of them and increase in their number coincides with tx fee spikes.
Looks like this is a job of a mining cartel as it tricks Bitcoin Core's fee estimation algorithm.