The only cache is the g-code model list. There we already compute the images. But we still need to reanalyse them on start since the gcode might have changed or settings for timing calculation. Only with the current setting recalculated they will get computed correctly. Also except copy from g-code all other methods have no image precomputed. Caching makes only sense when you always print same file with same settings, but that is often not the case.
