Uhm... I wasn't planning on investigating. I was just stating that I'd have to investigate if I wanted to pinpoint the changes, but even then it wouldn't tell us much.
You have to understand that Adobe doesn't publish the method they use to determine whether a page is actually Landscape or Portrait. It involves quite a bit of clever guesswork and, as evidenced by your issue, it isn't fail safe. So again, trying to determine if any change in our code or in their libraries, anywhere, impacts their detection algorithm is nearly impossible.
Just to give you an idea:
landscape does not necessarily mean a page's width is larger than its height. For instance, take a
portrait page and rotate it 90 degrees: does it become
portrait? Nope, it becomes
rotated portrait. But let's say the text on the original page was written sideways, does rotating it make it
landscape (because now the text becomes right side up)? Nope, it's
rotated portrait... but the Adobe Reader may well tell you that it's actually
landscape... or not (it depends on the percentage of sideways text vs. right-side-up text... plus some other clever detection parameters that we don't know about). And what if you had rotated that portrait page by
minus 90 degrees? Then would the page be considered
reverse rotated portrait or
reverse landscape??? ... ouch, my head hurts...

I'm afraid there's not much we can do about this.