Beauty lies in the eyes of the beholder
But 3D is hardly compareable to 2D because you see only one surface, in 2D this also wouldnt be such an eyecatcher (one line).
The first one has not a fixed iteration depth, the surface is estimated when the DE gets lower than a given stop value, independent from the iteration count. That is a very natural approach, i think.
I have rendered the last one again with higher accuracy, just wanted to know what happens: