First, what we're looking for:
This plot shows the X vs Y count of mouse input. This means it's basically just tracking your mouse movement. This is essentially a graphed version of the fast-swipe slow-return test for acceleration (like if our sensor was a pen on paper). If we see the the returning line overshoot the origin and go into the negative region, we have negative acceleration. If we see it not reach all the way to the origin, we're experiencing positive acceleration. Of course, there's a margin of error due to imperfect horizontal movement (Ey dude with the cool mouse rig, mind doing some mousetester plots? It'd look great and eliminate this variable!). My plots are 7,000 counts long and we see -4 count net difference and +20 count difference (explained below), meaning a max acceleration of 0.3%. That, to me, sounds like it's within the margin of error.
These are some plots I just took with my G203, I'll explain them a bit. All are at 600 DPI 1000Hz.
Here's a plot of the x/y of a G203, fast swipe right, slow return. If you look at the first image, it looks like there's a HUGE amount of acceleration present (look at where the other line leaves off, roughly 400 counts on the X axis).
If, however, you enable lines on the graph, you see more detail. Mousetester truncates the points to a lower resolution, so the last large dot isn't always the last sample. Showing lines includes the more fine details. I have no idea what the details of this behavior are, but I do know if you zoom, you can see more detail.
Zooming in (click and drag a region with the mouse wheel), we can see we horizontally overshoot the start point on the X axis (due to experimental error; it's a negligible -4 counts).
Second test, same sequence. First with the dots only, we see what looks like acceleration (albeit much
Enabling lines shows we're much closer to the origin.
Zooming in, we see a difference of roughly +20 counts. This is within margin of error considering a total 7,000 counts each way.
So two things I'd like to test: one, let's enable lines on the graphs so we can better see what's happening, and two, let's zoom in a bit so we can see the actual count difference. It's clear that my G203 is fine, but there might be a few things to explain what's happening. A preemptive guess I have; occasionally the G203 stutters and doesn't sample correctly. I'm not sure when this occurs (i.e. what movement speed/settings), but I recall seeing it before. I think I have evidence to support this theory, but unfortunately my sister is using my G203 right now so I can't check. Rather than acceleration, I think this might be some sort of polling stability issue? We'll have to use a different graph type to check that. This would be a much better explanation though, as it would explain why some people experience it and some don't. Additionally, such a thing could very possibly be affected by the max CPI bump from 6k->8k.
Edit: whoops, images in the wrong order.