Why a single percentage is usually misleading
"Card A is 15% faster than card B" is only true at one resolution, in one workload. Two cards with similar shader throughput but different memory bandwidth can be level at 1080p and separated by a wide margin at 4K, because bandwidth pressure grows with pixel count. That is why the table above breaks the comparison out by resolution rather than reducing it to one number.
Processor comparisons have the same problem in reverse. Two chips can be indentical in a graphics-limited game and far apart in a simulation title that hammers a single thread. Comparing across several games is the only honest way to do it.
When performance is not the deciding factor
If two parts land within a couple of percent, performance is not what should decide it. Look at memory capacity, which ages far less gracefully than raw speed; at power draw and the supply you already own; at physical size against your case; and at price today rather than at launch.