Why do deep Keras models become harder to optimize?

Depth increases gradient instability and optimization complexity.

Deep models suffer from vanishing/exploding gradients, poor conditioning, and internal covariate shift. Optimization landscapes become highly non-convex.