If the inductor is small enough so that its current never goes continuous, even at the heaviest load, then the RHP zero simply disappears.  This is because then any change in inductor current called for can be achieved in one cycle.

When the inductor is relatively large so that its current is heavily continuous, then the RHP zero arises because in order to increase the output, duty cycle must be increased to build up inductor current, but this also immediately results in a smaller portion of the slowly building inductor current being directed to the output (so output current falls at first) with the increasing portion being dumped to ground in order to apply positive net voltage across the inductor.

A system with one or more RHP zeros is known as a non-minimum phase system (anything with pure delay falls into this category) and cannot be compensated via global feedback for loop gain bandwidth much past the frequency of the lowest RHP zero.

The article you linked includes a derivation of the expression for the frequency of the RHP zero: fz = Ro/Lb*(1-D)^2.  This only holds if the inductor current is continuous where simple duty cycle averaged voltages drive its current.  If the inductor current goes discontinuous, then a third system state is introduced, which alters the system's two-state dynamics, completely eliminating the RHP zero.

If a RHP zero exists under certain line/load conditions, this may not be a limit to system dynamics.  It is only necessary that the RHP zero not be allowed to fall low below somewhere slightly above the switching frequency so that its phase effects don't come into play before the normal PWM limits to loop gain bandwidth.

Rather than work with equations, you can use the simulator to see how the RHP zero moves and various system parameters are stepped (e.g. inductor value, input voltage, output load, etc.).  This zero is easy to see by plotting the ratio of current out of the boost diode divided by current through the boost inductor.

A RHP zero in the transfer function is problematic because it is not possible to cancel it out with the compensation network.  The RHP zero in a boost converter can be dangerous because it moves around and, if not accounted for, can cause the converter to go unstable under heavy load where there is increased likelihood of failure.

Its mere existence is not the problem, but dealing inadequately with its possible effects might be.  Obtaining low output impedance beyond the general vicinity of the RHP zero cannot be done actively via feedback - it must be done passively via a large output capacitance (or some other means such as post regulation).  A large output capacitor can lower the size of load steps, but not the time to recovery.  Also, a larger output capacitor will subject the output to a larger energy dump if an overload collapses the output.

In a boost converter, if the current is discontinuous, a small increase in input on-time has the effect of increasing output on-time.  However, if the current is continuous, a small increase in input on-time directly subtracts from output on-time and can actually make the next lump of charge delivered to the output decrease even though the duty cycle is increasing (which must ultimately and eventually after a certain delay, result in more output current).  It is this momentary "backwards" action (i.e., delay in response) that is a sure sign of a looming RHP zero.

Do you know what the pole and zero pattern is that approximates an ideal delay?  If not, since it is a basic EE course item, just look it up on the net.  The bottom line is that RHP zeros always are associated with nonrecoverable delay.  When coupled properly with symmetrically corresponding LHP poles, their effect on magnitude is exactly cancelled (resulting in a perfectly flat magnitude response, but with crashing phase - that is, an ideal delay).

Hope this helps.