Corrected my post above. Got a 4 where there should have been a 2 and since it was an exponent, it really made a difference.
If these are groups fired with two systems, one is more accurate than the other in spite of having the same average.
To have any chance of providing a reasonable answer when combining accuracy figures, the figures need to actually be representative accuracy figures.
So if you're getting a wide variance across your groups (as with the first system in your example)--especially with a relatively small number of groups then you won't be able to provide a representative accuracy figure and the combination won't be representative either.