suppose we take our "0 angle" to be the (positive) y-axis (i think this is what you are saying).

in this case, if φ is the angle between the point (x,y) and the y-axis:

x = rcos(φ + π/2) = rcos(π/2 - (-φ)) = rsin(-φ) = -rsin(φ)

y = rsin(φ + π/2) = rsin(π/2 - (-φ)) = rcos(-φ) = rcos(φ)

after rotating through an angle of θ, our new coordinates are:

w_{1}= -rsin(φ+θ)

w_{2}= rcos(φ+θ)

w_{3}= z (the z-axis is unaffected by a rotation in the xy-plane).

using the SAME trig identities as before, we have:

w_{1}= xsin(θ) - ycos(θ)

w_{2}= ycos(θ) + xsin(θ) = xsin(θ) + ycos(θ)

so, no...it doesn't matter "which axis" we count our "rotation from".

if you think about it, the formula for a cone:

z^{2}= x^{2}+y^{2}, is the same even if we "switch the x's and y's".

in fact, you could start measuring "the original angle of the point (x,y)" from ANY line in the xy-plane passing through the origin, and the formula for the w's would STILL come out the same, but the algebra gets more complicated (because the formula for x and y would be defined in terms of an angle from a line which is itself at some angle from the x-axis).

put another way, ALL diameters of a circle (and hence of a "slice of a cone parallel to the xy-plane") work just as well, so why not pick a "convenient" one.