Why does increasing model depth sometimes reduce Keras performance?

Deeper models may suffer from vanishing gradients, overfitting, or optimization difficulty.

While deeper networks increase representational power, they are harder to optimize. Without residual connections or normalization, performance can degrade.