The index write dedupe code checks if any buckets are missing, and if so writes all buckets.
This means that, when a series continues from one day to the next, when writing the chunk that spans the day boundary all the previous day's entries are written again.
It may be this way because the interfaces don't talk about buckets, but this seems like a significant win for those of us who are iops limited.