Dynamic parallelism vs flat kernels

Accelerated Computing CUDA CUDA Programming and Performance
1

Generally speaking, when can dynamic parallelism out-perform a (well-written) flat Kernel?