Very quick info on relative analog (mouse) movements:
I'm calling a relative change of one as a single "count", not "pulse" as mentioned in prior posts.
The "*"ed stuff have some exceptions to that rule, so please keep that in mind.
Modern PC mice get 4 counts per tooth* (the two edges per tooth, times the two sensors).
Original arcade machines could have gotten 1, 2 or 4 counts per tooth. The only one I'm sure of is 720, and it gets 2 counts per tooth.
What is now done in the mouse firmware was done on the arcade PCB*. This means mame cannot emulate this part of the hardware*.
Different PC mice have different counts per rotation.
Windows OS mouse sensitivity, acceleration, & speed are not supposed to effect directX games according to MS, but for some mice it does.
The above might be because some mice can scale the output in the firmware. The most common is a simple logrithmic pattern: instead of 1, 2, 3, 4 counts per communication, it sends 1, 2, 4, 8 but it could be anything.
The most reliable* way to test the true output count of a PC mouse device in windows is set the mouse speed, sensitivity, enhancement, acceleration, etc to the lowest possible or none. Otherwise the count could be modified by the firmware, OS kernel, or driver.
Most* Windows people like to have an acceleration (called "enhanced pointer precision" in winXP), which scales the mouse much like the firmware scaling, but usually with weirder curves. (Such as 1, 1, 3, 6, 10, 16, 24, 24, 24 ... instead of speed of 1, 2, 3...)
To sum up:
Matching physical tooth count is not good enough, nor always needed.
You could have the orignal hardware hooked up to a PC, and even with all the speed etc set to plain, have the game see 2x or 4x the count per revolution. With the windows settings to default or normal PC values, it might be perfect at low speeds, and totally off at high speeds at the best.
And spinners with lower tooth counts per wheel can (however unlikely) exactly match (in mame) the original hardware count if, for example, the original hardware was a 1x count per tooth with 4x the teeth (and all the windows settings are perfect).
FWIW:
Modern PC mice use "state change detection" of "quadrature signals" to see it turn. "Quadrature singals" is different than "pulse singals", but the output to PCs of either would be the same. I doubt any arcade hardware used pulse singals. The old tech was "edge detection" (of the same quadrature singals), and this was very common in the early arcades. None of this matters to counts per revolution matching.