Do you have any benchmark showing it's really getting slower on large data sets?
All the operations here are supposed to be fast operations, but perhaps a better feature would be to add a new -a switch to simply appends the data, binvar or not.
/bunset is as fast as you can get, a switch in /hadd unsetting a binvar really seems like cosmetic, but a way to append would avoid a call to $hget() + /bcopy + /hadd in the case of a binvar usage, which would be nicer indeed.
