Changing neighbor distance and number of triangles - increases the simulation time?

Hello,
I am trying to speed up an existing simulation and have reduced the number of triangles in the STL - from about 3000 to 250 -. This saved me about 50% of the total time. Then, using the original STL, I reduced the neighbor command to 0.002, since in the previous simulation it was 0.004 and the number of average neighbors was very high. This also gave me a reduction in total time of about 50%. After these findings I thought that a combination of the reduced STL and the smaller neighbor distance should reduce the time even more. However, it turned out to be just the opposite. The time increased. Does anyone know what this could be? Shouldn't the reduction of the triangles make the simulation faster because there are less points/vectors? Especially when mixing later where I move the STL I would have expected this. Or do I understand something wrong?

To understand, the Y-axis in my graphs is the time per step and the X-axis is the number of output points on the screen from the Liggghts log file. The first five points are the filling of the particles and the remaining 400 points are the mixing.