one interesting point that came up at the bitdevs in austin today that favors remove that i believe is new to this discussion (it was new to me):

the argument can be reduced to:

- dust limit is a per-node relay policy.

- it is rational for miners to mine dust outputs given their cost of maintenance (storing the output potentially forever) is lower than their immediate reward in fees.

- if txn relaying nodes censor something that a miner would mine, users will seek a private/direct relay to the miner and vice versa.

- if direct relay to miner becomes popular, it is both bad for privacy and decentralization.

- therefore the dust limit, should there be demand to create dust at prevailing mempool feerates, causes an incentive to increase network centralization (immediately)

the tradeoff is if a short term immediate incentive to promote network centralization is better or worse than a long term node operator overhead.

///////////////////

my take is that:

1) having a dust limit is worse since we'd rather not have an incentive to produce or roll out centralizing software, whereas not having a dust limit creates an mild incentive for node operators to improve utreexo decentralizing software.

2) it's hard to quantify the magnitude of the incentives, which does matter.