PNG-s are far from ideal to run as content.
When decoding and uploading of the png is not finished before the next render pass, the timing of the clip will be off compared to the other clips.
In your top layer you try to display 100 pngs a second with the auto pilot. That would need 100FPS playback so they are all visible, but the composition FPS you get is 22. Auto pilot is evaluated at every render pass, you get 22 render passes a second, so you get lots of missed auto pilot actions because of the much higher rate you expect. That's why your layers go out of sync.
Loading the 0-9 pngs as 4x image sequence clips and switching off auto pilot, already gets me back to 60 FPS on my macbook with quite perfect timing: https://resolume.com/support/en/stills#image-sequences