Hi Kyle,
It seems to me that the 1st seven rows of your original glt matrix would be good enough to catch what you want: the voxels where a difference exists between at least two conditions. That is,
A B C D E F G H
_______________
1 -1 0 0 0 0 0 0
1 0 -1 0 0 0 0 0
1 0 0 -1 0 0 0 0
1 0 0 0 -1 0 0 0
1 0 0 0 0 -1 0 0
1 0 0 0 0 0 -1 0
1 0 0 0 0 0 0 -1
The reason is this. Any other pair-wise constrast is essentially represented by two of the contrasts in the above matrix. For example, the contrast in row number 8 is basically expressed by the first two rows. If we denote the corresponding null hypothesis for contrast B-C as
H_01: B = C
and the null hypothesis for the first two rows as
H_02: A = B, and A = C
If you reject H_01 (B is very different from C), either A is very different from C, or B is significantly different from C (it can't be neither), so you would reject H_02. So H_01 is already included in H_02, and you won't miss anything.
Needless to say, such a glt matrix is not unique, and you can come up with several other matrices with seven rows. The pivotal point here is that those rows have to be independent with each other.
Gang