Using a simple passthough shader that just samples a texture and writes to an output image, here are my results.
Framebuffer size is 1280x720
GPU: GEForce 1660
Optimal compute batchsize: 8x8
Optimal compute number of workers: 14400
1 pass:
Compute
FPS: 600
Fragment
FPS: 770
10 passes:
Compute
FPS: 180
Fragment
FPS: 670