Part of the problem with the curation algorithm is that it's fairly strongly limited by the witnesses' calculation power and the number of blockchain operations necessary. There are ways that it could be improved, if those limits were reduced. Beyond that, it could also be improved by competition: if multiple entities and multiple priorities were driving the development of various curation algorithms, they would advance our knowledge of how to algorithmically reward the discovery of valuable content.
Both of these things can be achieved by decentralizing the calculation and distribution of curation, moving responsibility for it from the core blockchain to the posting interfaces (Steemit, 
An interface would take a beneficiary payment, and run their own, off-chain calculation to determine the payouts for the voters, then send the rewards via transfer. This offers several advantages over any system with a set percentage of curation on the blockchain level:
- Significantly reduce the processing power required by witnesses by offloading the curation calculation.
- At its base level, this would maintain the level of blockchain operations, or increase them by 1/post for interfaces that don't already take a beneficiary.
- Interfaces could choose to batch payouts for their users; a daily or weekly curation payment transfer rather than one for every vote would reduce blockchain operations.
- Allow for more complex calculation of unexpected discovery via methods like Bayesian Surprise, which is impractical on-chain.
- Each interface could choose its own curation percentage, allowing both posters and voters multiple options.
- One or more interfaces could implement the user-chosen curation percentage proposal without increasing blockchain load.
- Users with direct-access skills could choose their own curation percentage, though they'd be responsible for distributing the rewards.
- Each interface could develop its own algorithm based on its own priorities.
- Batching transfers would eliminate the need for dust curation; small users could accumulate small ledger amounts until they reached the minimum transfer amount.
- Competition between algorithms would lead to much faster development of our understanding of how rewards are optimally distributed.
This is sort of a hacked-up SMT-esque system that can be built on top of the current Steem today, as a proof-of-concept without any consensus change required. Many of the blockchain benefits depend on removal of the existing algorithm, but if the proof-of concept is demonstrated to be practical and effective, the base curation layer can be removed at a later date.
I didn't invent this, DTube did. I just noticed its potential to satisfy broad categories of stakeholders in the debate over curation rewards. 50/50 can happen if you want it to, so can 69/31, so can 100/0 if you like. User behavior can determine which are successful.
