Are More Layers Beneficial to Graph Transformers?