Yep, this annoys me mightily too but I've always thought of it as a web-design fail rather than a browser fail.

I was about to ask how the new Firefox knows what constitutes a "fixed header" and what doesn't - like, what if the overlay only goes part way across the page, or something? Then I read at the above link that they've drawn up some clever heuristics to guess the 'right' scroll amount in each case.

"It works well on many pages. No doubt there are some pages it doesn't work on. It seems like a good idea to start conservatively here."

I cautiously welcome this 'fix' for my own casual web browsing, while fearing for the rare trouble it might cause a responsible web dev who might have a legitimate need for a particular kind of overlay while having page up/down still work in the currently established way. No such need springs to mind but I'm sure that if I thought about it for long enough, in this age of games in the browser and whatnot... (some sort of fancy web comic perhaps?). Now needing to workaround what is more complex and unpredictable behaviour, this is the kind of push/pull between web developers and browser implementers which a few years down the line could turn into a bloody impossible mess for all concerned.