Why does motion planning use linear interpolation instead of b-splines?

It's a question that I've been asking myself for quite a while. Linear interpolation is quite limited in its accuracy as well as its ability to maintain constant velocity. Have there been any attempts to replace that with higher order polynomials / b-splines / NURBS etc? I saw the recent great achievement to implement s-curve acceleration (add a setting for jerk), but here I'm thinking more about directly improving the motion planner. I understand that this is significantly more resource intensive, but honestly, for a modern PC it should be trivial to compute.

I think the main thing here is that there are little real benefits of higher order
interpolation, at least for the speeds, accelerations, and servo period used on
typical LinuxCNC machines.

For example, using a relatively high performance metal cutting machine with say
100 mm/s cutting speed and 0.5G acceleration, the chord error caused by linear
interpolation at the highest centripetal acceleration and at a 1 ms servo thread,
is only about 0.6 microns.

The error just gets smaller a lower speeds and accelerations.