This behavior that index scans will flip into sequential ones once they access enough of a table is one of the more surprising ones to many people. Random disk access is so much slower than sequential that it makes sense to do so. But if all your data is cached in RAM (the case on examples shown here), the advantages aren't as obvious. And that may very well always be the case with your data. In that case, you could even consider lowering database query parameters like random_page_cost to make the optimizer understand your true situation.
This whole area is one of the things to be wary of when executing your own queries that expect to use indexes. You can test them out using some subset of your data, or perhaps using ...
