In short, the kernel VM system when dealing with memory pressure scans through in-use pages and tries to keep them in a balance between active and inactive markings. The inactive pages are scanned for reuse while marked as inactive. If they have been reused, they are marked as active and some other page must move from active to inactive state to detect if it is in active use. So, inactive is a misnomer. It is shorthand for "possibly inactive, lets try to verify that."
As you discovered, the internal balance we (currently) strive for is approximately 2/3 active vs 1/3 inactive (although the algorithm is more complicated than that, it serves as a general approximation). In addition, many things can force these out of balance, and if the VM system isn't under memory pressure it lets these drift. It's only when there is pressure that you tend to see them numbers get to a pretty-close-to-fixed ratio. At that point, it's not about the ratio of the numbers, but the specific pages tagged with each classification, and their accesses, that matters.