Which is better to repeat trial within run or across run?

Hi all, I have read some awesome works aimed at using encoding model to decode natural images from BOLD signals, like kay,2008.

In this paper, the identification run use 12 images to repeat 13 times. While the whole identification images have 120 images, then subjects will participant 10 runs.

I am wondering whether the changes of subjects’ states along the session of 10 runs can influence the decoding performance? If exists, does it make decoding easier or harder?

For a view from psychology student, I think it’s better to repeat images across runs, that means subject view all the identification images in each run. But will it bring more random effects?

Thus, my question is: to get a more accurate representation of single image, which one of repeating designs above is better?