I think some people are both right but there is need to understand what the word "Tolerance" actually means in engineering terms because it would shine a light on what the regs say.
A tolerance is actually defined as a range of values around a mean.
A tolerance of 10mm plus or minus one mm means that the intended value is in the range 9mm to 11 mm.
Ideally, the average value is 10 mm.
If CASA directs that a tolerance of Two miles be applied, without specifying any limits, then one could be forgiven for assuming that if your average track misses the control boundary by Two miles, then you are complying.
If you wander all over the sky like me, then you are still complying as long as the closest approach of your average course does not take you closer than Two miles. In other words, your closest approach might be one mile or less provided your course also deviates two miles in the other direction at times.
If CASA or ASA wants to specify that no VFR pilot is permitted to approach within Two miles of the boundary of a control zone then they should effing well say so. As in 2 miles minus zero miles plus whatever miles.
In that case you would plan for a course: Two miles plus another mile or Two (your own expected variance) away from the zone.
To put that another way, the regulations and the courts need to understand what the word "tolerance" actually means otherwise they should be written using the word "limit".
The issue of GPS usage raises another issue - risk shifting. this is planning to use tighter tolerances because you think your GPS is more accurate than Mk1 eyeball. The end result of that is what is called "GPS assisted collision" (eg Brazil).
For example I have had another aircraft flying exactly overhead on a reciprocal course in the Melbourne VFR corridor (sugarloaf to kilmore gap waypoints) - which is one reason I always offset my tracks and "fly all over the sky".