Windows

The other possibility here is that Statcounter is trash.

People are taking its findings as if they're gospel yet it is a very crude way of measuring how many users there are for an OS. It basically just measures how many users of each OS it sees each month - it is very difficult to consistently correct that data month to month, or match the same data to the same user each month. Lots of people use ad blockers and other privacy tools in their browsers which could easily break the sort of tracking Statcounter relies on.

Essentially, flucations in their data may just reflect the poor accuracy of their data rather than actual swings of 1%.