I guess the reason is that large transactions are more likely to conflict with something in the mempool and are therefore much harder to push to the miners, correct?
Yes. Smaller txs have more overhead, but this is still more efficient than most of the sweeper spam that creates utxos and uses big sigs. Ideally the miners would OP_RETURN it themselves in big txs but it seems they were too lazy.
Does this mean that the keys for the remaining 172 BTC will be published later?
My guess is that coinwallet overestimated the impact of the "giveaway", so gave up. The problem is, of course, that one can only construct low-fee txs with dust inputs, which has no real impact on the network (apart from artificially inflating the average blocksize). Normal-fee traffic was unaffected.
I guess coinwallet will consolidate the remaining dust outputs themselves at some point since there is no real point in giving it away. But if they publish more keys then our bots are ready...