I think it would be neat to have - but personally I think there are higher priorities (asynchronous IO, faster executor, better buffer mapping, different connection model, built in automatic failover). And I don't think any of my colleagues was planning to work on the topic. A good implementation would be a lot of work, with an uncertain outcome at best.

There have been a good number of efficiency improvements in the last releases to postgres' btree implementation in the last few releases. Not the same as what you're looking for, obviously, but it might still help.