I think in the long run they are both 100% . I don't know about GPS but DCF77 only updates once a minute on a perfect signal for the other time your are relying on the Arduino for timing.
Yes, I looked into the DCF77 signal a few months ago because I had an antenna from a broken clock. I decided not to use it because the signal is terrible in this room (I live in a tall building). The signal was almost 100% noise. It was OK out in the street but not in here.
(It's fun to look at the DCF signal on an oscilloscope - you can see the pulses very easily out in the street...indoors it was just garbage, no clear pulses at all)
My clock looses a second around 6 or 7 times a day (it always corrects itself) so I presume using the Arduino for a clock without any external correction is not a good idea.
It would be terrible. If you put two Arduinos running "blink" side by side and simultaneously press reset on both they only take a minute or two to noticeably go out of sync with each other.
I would expect 1 second in 2 weeks on a properly adjusted quartz clock in a reasonably stable temperature. Maybe I could add a trimmer to the Arduino crystal ?
Maybe better to figure out the time error in software (you have a good reference clock!) and add extra cycles to the calculations to compensate.