Peter, you're confusing me again.
Its probably because I tend to use a very specific case of interpolation, that being a linear interpolation between two points. The concept of circular interpolation has always seemed odd to me.
Given a multi-axis linear vector move, the position path for each axis between the two points can still be a known function. If you define the move in terms of the vector path the motion of the orthogonal components is a known relationship to the resultant vector in terms of position and all it's derivatives. If you are trying to optimize the move you may need to back-calculate the available peak velocity, acceleration and jerk of each axis. If you simply use the peak values from the vector move and don't allow a single axis vector move (on one of the planes) to violate the peak capability of an axis you have a simple, constant limit.
Similarly, with a circular move you only have non-zero initial velocity and acceleration if you start with a non-zero omega for the vector. If you accelerate the vector rotation the component axes will also accelerate and never at a greater value than the tangential acceleration at the end of the vector.
It all comes down to profiling.
Keith