None of those benchmarks really add up to much hard information. Running at legacy 13" resolution defeats the purpose testing any high resolution card optimized for the reason for its purchase. Nobody bought one of those monster CRT/VidCard combos to run at 13" resolution, so no point in doing much if any optimization for tiny displays.
Even testing at 1024x768 probably won't do much good for high end cards optimized for TPD thru 1600-1200.
So, what the MacBench publishing graphics test are measuring is: How fast can a machine and its graphics card execute QuickDraw calls. You can test that speed factor at any resolution.
Measuring that at "low" resolutions like 640x480 or 1152x870 is still useful. The other MacBench tests will product numbers at any resolution, but they're more about specific tasks/functions like fill rates, tesselation, etc etc.
FWIW here, 640x480 was common for most day to day users into the mid-late '90s, and it was common for second-hand computer buyers not to have access to particularly large displays. (Although that got better into the early 2000s and mostly it was a matter of finding adapters, especially for people a bit older than I was who had slightly more discretionary income.) (For my part, there were high end Quadras for days at the local used computer shop, but the only Apple monitors you could get were 13/14-inchers.)
To the greater point: resolutions higher than 1152x870 didn't become "normal" until way later. The PCI PowerMacs are the first Apple graphics solution I'm aware of to support over 1152x870 and the Beige G3 in 1997 is the first Mac to support over 1280x1024 (up to 1600x1200 or 1920x1080 on the G3's Rage, if you put in the max vram.)
I'll have to look at some 1993-1994 magazines for this but it would be interesting to see how important performance is at all in the context of "getting to 1600x1200". This sort of hearkens back to a discussion we had before where we found out 24-bit color in 1991 pretty much only existed at 640x480 and any higher resolution/larger displays were typically 256 colors or monochrome. I think jessenator mentioned in another thread that publishing workflows tended to have "the layout machine" and "the color work machine" and so you'd just have different computers for those different tasks.
The other thing w/re graphics benchmarks is, to the extent possible, all solutions that are being compared should be run in the same machines with the same configurations and software loadouts. Especially for unaccelerated cards, the speed of the computer can have a huge impact because quickdraw will be rendered on the CPU. An 8100 with a G3/400 upgrade will probably win at every graphics test at every resolution, even if the card involved is, like, a Toby. (I mean, on that card the resolution will probably be 640x480, but.)