Good morning ZmnSCPxj,

The issues you point out are indeed important to note. Thank you for your wonderful feedback!

* There is a practical limit to the number of UTXOs you would be willing to receive in the swap.
  * Every UTXO you receive increases the potential fee you have to pay to spend them, meaning you would strongly dislike receiving 100 UTXOs that sum up to 1mBTC.
Absolutely agree. It wouldn't be particularly nice to have to manage that.

  * Thus, a practical blockchain analyst can bound the size of the sets involved, and the problem becomes less than NP in practice.
Definitely, though they first have to consider all subsets of a fixed size with values bounded above by the value of the unknown sum. So the analyst has to search through all fixed size sets (up to the practical bound) whose elements are less than a maximum sum. This is a number of choices that is (in a crude estimation) exponential (in the size of the UTXO set), and polynomial in the number UTXOs below that maximum sum value on-chain which can be pretty big at sufficiently large value-transfers.

* If you have a single UTXO and split it, then swap, anyone looking at the history can conjecture that the split involved is part of a CoinSwap.
  * The split is now a hint on how the subset sums can be tried.
You're right that anybody could conjecture that it is involved in a CoinSwap, however in my proposed protocol the swap would like a (schnorr) P2PKH to the chain so you'd have to make that conjecture for every UTXO, so it's not much of a hint. Especially so noting that one, both or none of the outputs could be part of a swap.

* If after the CoinSwap you spend the UTXOs you received in a single transaction, then you just published the solution to the subset sum for your adversary.
  * This ties in even further to the "practical limit on the number of UTXOs".
    * Because it is not safe to spend the UTXOs from a single CoinSwap together, you want to have fewer, larger UTXOs for more flexibility in spending later.
Yes, this is definitely a weakness and some over-the-top UTXO management techniques (e.g. try to avoid combining different UTXOs in a known set into the same transaction by default, where possible) would be needed or like you say fewer larger UTXOs.

It's interesting to note one can pick some subset of recent UTXOs and add up their output values, and select that as the amount of value transfer to exchange in a given operation. Resulting in a bit of added obfuscation as there are now seemingly (at least) 3 utxo sets that add up to similar or identical values, but only two of which are really participating in the swap.

I believe belcher and waxwing and nopara73 have been working far longer on privacy tech, and you should try to get in contact with them as well, they may know of other issues (or solutions to the above problems).
Thank you for your input and suggestions! I will reach out to them.
 
--
Germán
Mathematician