I think that’s a good idea. Dust Expiry should reflect a balanced principle: while every UTXO imposes a cost on the system, non-dust UTXOs also provide value—since the holder’s conviction contributes to Bitcoin’s overall worth. The mechanism should penalize only those UTXOs so small that their cost clearly outweighs their contribution.
Do you mean an accumulator scheme that allows state updates using an inclusion proof? Otherwise, if updating requires knowledge of the entire set, the accumulator introduces unnecessary overhead.
Also, the set of unspent expired UTXOs isn’t actually that large: location pointers of the form
pointer = (block_height, tx_index, output_index)
can be naively encoded in under 8 bytes. That amounts to roughly 800 MB for the ~100 million spam UTXOs created in recent years.
Moreover, the set can be compressed significantly. First, since the order doesn’t matter, you can sort the entries to reduce entropy. Second, both block_height and output_index have quite low entropy and can be compressed efficiently. So you can likely get away with 4 bytes per entry. Perhaps even less than 2 bytes.
And since this cleanup mechanism disincentivizes UTXO spam attacks, the set likely stops growing so rapidly.