I am actually using MadCAM. It works pretty well for what I do.
I have been doing a fair bit of reading scholarly articles on NURBS machining. Mainly this is driven by the trajectory planning issue - lots of papers are being written about it. Clearly defining the path as a NURBS curve lends itself more to good trajectory planning than defining the path as a bunch of connected segments.
I emailed Joakim and he is considering adding NURBS toolpath output to the next version. This would be a huge step forward, because none of the other low-end CAM programs can generate that code, which seems to be limiting the developers.
Of course there don't seem to be many controllers that will accept the NURBS gcode at the moment, but that is changing. EMC has a biarc interpolation method in 2D now apparently, and some other guys claim to have it working in 3D. But I haven't seen any video, other than a short one by Martin Duzi in 5 axis EMC, and he is using his own HAL interpolator to generate the toolpath from a regular line/arc toolpath.
In the interim, I think I need to mess around some more with the CV settings and the Tempest planner. When doing concentric surfacing paths, for instance, the jog from the end of one lap out to the start of the next one is too steep, which sets up resonance. There has to be a way to slow that movement down without slowing the rest of the toolpath.