Very interesting that it is Hex elements, as I visualized it with tets seeing these “fingers”. What do pressure and velocity contours look like over the same area? I am a bit suprised that it takes 8 cells to go from 1 to 0. A sharp interface is expected, correct? What timestep are you using and after how many timestep is this? A part of me wants to say this could be errors building up from solution that does not converge well each timestep. What does the convergence look like?

One last thing I would try is decreasing the porous media resistance to see if they still develop in the same way.

The cheat would be to look at a quasi-2D model. Then you guaranteed not to get these ;)

EDIT: And how are you specifying the PM resistance?