Mostly because it probably doesn't hurt anything to have those instructions still there, or at least the cost of having to deal with the lone enterprise customers that still use the instruction in their legacy apps is bigger.
There are a whole lot of gates available for use: every time you shrink the processor you make space for more. I bet those old unused instructions don't really occupy a lot of them, and perhaps some of them are only implemented in microcode in terms of actually used instructions. There is no pressure to make semi-dead instructions fast.
I am not an expert here but I'm semi-confident my answer is accurate. Please correct me if I'm wrong.