Disclaimer: I can only speak in reference to PhD graduate work. I don't know how masters programs work since masters programs in my field are virtually non-existent.
The problem that I have with counting graduate GPA is that there is alot of filler in your graduate GPA. I get a grade for my research hours every semester. Most semesters that is like 13 hours of research and it is an "A" every time. That grade is highly subjective. Sure I am a good student and work hard and produce alot of data, but how do you grade something like that. More importantly, it completely drowns out any coursework. I think for my PhD work, I have 10 research hours for every 1 coursework hour. I don't value an "A" in my research hours as much as I value an "A" in an actual graduate course, which is much more difficult. Therefore, I don't put much stock in graduate GPAs. I can also tell you for sure that if/when I seek out a fellowship or post-doc to further my research training, they will not even look at my graduate GPA. Graduate GPA (in my field) has too much filler in there to be an accurate reflection of your performance in graduate level courses.
I think the advantage of doing a masters would be proving yourself, especially if you do a research-intensive masters. Doing this coursework will not get you past any GPA screens/cut-offs at most schools since most look at undergraduate GPA only. Like I have said before, I think advanced degrees can give you an edge (if they are in line with your career goals), if your numbers are competitive.
For the record, I know Penn does not count graduate GPA as part of your last 45. It is your last 45 undergraduate.
