Miracle of PBC and Meshes

I needed to simulate a simple troughed chain conveyor, so I made a box from -1 to 1 and set PBC in conveying direction.
After the first plate goes through the PBC it starts to flicker, this means it is one time at the end and the other time at the beginning of the simulation box.

Furthermore simulation time increases by a factor of ten after that step (in the example it happens between step 44.000 and 45.000).
BUT (and now the magic happens) when I start a fresh run after let's say 65.000 iterations for some reasons, the simulation time becomes normal until the next mesh reaches the boarder.
3rd wonder: It takes more time when I devide the mesh in 8 plates á 2 triangles instead of having all plates in one mesh (1x16 triangles)

The timing tells me, that it is the modify time which significantly increases. I tried to use a profiler application (zoom in this case) but it only pointed me to Neighbour::bin_atoms() line 1980.

simple testcase attached (first 65.000 iterations bad, 2nd 65.000 iterations perfect, last 65.000 iterations horrible performance)

Tested with LIGGGHTS 3.1.0 git commit b3030a8fd1f722c77f69150c7b2d24b386eab848

Have a Nice Weekend,
Christian.