Image classification model (ResNet9)

Is it necessary that a model for image classification will always perform better if we used RenNet9 architecture or it can sometimes perform better without it.
Provided we used data normalisation and same number of nn layers in both of them.