If I'm not mistaken, the current wallet randomly selects outputs for mix-in from the TXO set using a certain probability density function.
Is this behaviour somehow supported by protocol-level incentives, other than maximizing transaction privacy (e.g. fees)? Or might we, in the future, see wallets which constantly (re-)use certain outputs in their ring signature for performance or other reasons? Could such wallets undermine the mandatory mix-in rules?
There's no direct incentive scheme at the present moment.
The only protocol requirement is for the number of decoy outputs in the ring signature (and that they're of the same denomination, for non-RCT TX-es). Which ones get chosen is up to the wallet.
If someone were to use a different one, he would probably be fingerprinting his TX-es in some way. It would likely not be possible to exactly be sure which ones because of some false-positives from other wallets, but still - some could be marked as "likely origination from xx wallet implementation". If enough people used the xx wallet, each user would still be hidden among other users of xx wallet.
Considering that with RCT you will be able to pick any random output on the blockchain as decoy, it doesn't make sense to re-use same ones. At the end, you'd be sacrificing your privacy - for what? So, there's that indirect incentive to stick to the default.
I'm sure that this aspect will attract more research as monero grows, which will provide more substantial answer to your question. Some research has already been done, see MRL-0001 and MRL-0004.
