There are lots of scores, of CPU rating of around 10 or 9. And SpeedSys says that NPU is also present in most these cases.
In those results, they're also using a 486DLC. That's not always apparent in the screenshots, since Speedsys doesn't seem to properly recognize the DLC chips. And it makes sense that 486DLC+NPU would be faster than 386DX+NPU, because the DLC is a faster chip.
A 486DLC with an NPU scores ~10 (as compared to 6-ish for a 386DX+NPU), but if you pull the copro it'll score ~20 (versus 9-ish for a 386DX without NPU). I suppose you could say that the non-NPU scores are inflated, but I don't think anybody really knows what metric they use to come up with the numbers, so there could be some sort of reasonable logic behind it, I dunno.
I just could not imagine so many people had systems without FPU, to have scores around 10.
Really the only reason to have an NPU in a 386 is just for shits and giggles, to fill the socket. There's not much software that actually takes advantage of it... I think 2 or 3 games, and then CAD and business applications. That's about all. So it's not at all uncommon to have a 386 system without a coprocessor.
What I'm thinking about, the later boards can 'enable internal cache' on Cyrix DLC and it could give them the edge in performance.
The SiS boards are later boards. I don't know if all boards with that chipset support the DLC's internal cache, but I know mine does. And even if a board doesn't have native support for the cache, you can still enable it with a software utility.