That is a pretty wild swing. Something is changing. You can overlap targets to get a larger sample if you haven't changed anything. 3-shot groups by themselves can be rather variable, so unless they repeat, there is a good bit of luck-of-the-draw with them. Below is a computer simulation making groups with a random number generator that has its output constrained to produce a normal distribution on both the horizontal and vertical axes. This set contains something close to both the extremes of a 95% confidence limit for 3-shot groups, which makes it a bit of an outlier. I had to run the simulation a number of times to get an example this extreme, but you get the idea from it of what can happen at times and how drawing a conclusion about how the gun is performing is difficult with just any 3 of the nine shots. But combined, they give you a much better idea. Even if all the three-shot groups had been tight, when they combined, the group would be bigger because you can see that not only did the groups differ in size, the little red x's that mark the center of each three-shot group moved around. That center moving has a standard deviation equal to the standard error (sd divided by √n) on each axis.
Per the cleaning, I used to use Butches and other ammoniated cleaners, but the water-based chelating cleaners work so much faster and better for copper, it's a night and day comparison. This old article may be of interest.
Per the cleaning, I used to use Butches and other ammoniated cleaners, but the water-based chelating cleaners work so much faster and better for copper, it's a night and day comparison. This old article may be of interest.