I’ve inspected the distribution of documents per bucket, and found out that the distribution was poor - about 10-60 documents per bucket which resulted in a large number of buckets, which in turn might be adding additional time for unpacking of buckets.
Then I made a copy of the data in another collection, which has a granulation of hours instead of seconds.
I’ve also checked the document distribution over the buckets of the queried meta data.
Came out that 95 buckets out of 99 in total (which are in concern of the query) now have at least 900 documents (the bucket auto-close is on 1K).
Made the same queries and guess what?
Again avg. of 7seconds.