Why just so a small number of layers and filters is used in Graph Neural Network?

Question

I am interested in the successful application of Graph Convolution Neural Networks. For this purpose, I have investigated some papers and implementations and in summary, the