The text is working backwards from assumptions about oscillation occuring and finding the necessary equation, if so. Unfortunately, it's also using \$n\$ in two contexts. (One where it is used mathematically in the form of \$n\cdot 2\pi\$ and means one thing; and, another where it is used to mean the odd number of stages in the oscillator as applied in \$\frac{1}{2~ n~ T_d}\$ and means a different thing.) So you need to parse out the details a bit.
I'll use \$n\$ in the mathematical sense used in \$n\cdot 2\pi\$, so that this is only stating the obvious: \$\left(n\cdot 2\pi\right)~\operatorname{mod}~\left(2\pi\right) = 0\$, for all integer \$n\$. Which is only about the criteria that needs to be met for oscillation and is pretty much obvious. I'll use \$k\$ to mean the odd number of stages used. (It must be odd if there are to be two passes through the system.)
Given an odd number \$k\$ inverters, one pass through the chain ultimately takes the input and presents it at the output, inverted. So it clearly will take two such passes to meet the \$0^\circ\$ criterion. So the signal must pass through \$2\cdot k\$ stages to meet that criterion. However, it takes \$T_d\$ time per stage (the delay through the stage.) So that means the total time must be \$2~k~T_d\$. That means the frequency is \$\frac{1}{2~k~T_d}\$, by definition.
In general, it is noise that will cause it to oscillate. (You can perturb the simulation using the .IC for initial conditions to create it, so that the simulator doesn't find some middling quiescent point of stability that you don't want to see.) Noise at those frequencies which cannot make it around in exactly \$2\pi\$ will die out. Noise at frequencies which can make it around, just right, will be amplified and may survive for another go-around through \$2~k\$ stages, again (by which time the signal will already be clipped and the following gain will become 1, most likely.) Also in general, the more stages you use the less startup time is required.
There will be jitter in the process. The stages won't all have the exact same \$T_d\$ value and it won't always be the exact same in any one of them through all time, either. They aren't even driving all the same loads (as you will take the output from one of them.) So, expect phase jitter to occur and that it will be both temperature and time dependent.
It is surprising, but the second graph shows the inverting gain - measured with a test signal at the inverting input (because the phase shift is -180deg for low frequencies, including DC). This is rather uncommon but creates no problem at all (normally, the non-inv. gain is given with 0 deg. for low frequencies). So - this difference does not mean anything.
More important is the phase shift at the frequency for unity gain. This value determines the distance of the loop gain phase from the critical value (in case of unity gain feedback). This critical value is -180deg for the 1st diagram and 0 deg for the 2nd diagram. This difference is called minimum PHASE MARGIN PM. For PM=0 the circuit with unity gain feedback is unstable.
In the first diagram we have app. PM=(180-100)=80 deg. and from the 2nd diagram we derive PM=(45-0)=45 deg.
Hence, as far as the phase margin (stability margin against self-oscillations) for unity gain feedback is concerned, the first device is better (more stability margin).
Best Answer
No - it is not the classical phase-shift oscillator. The resistor from the inv. input to ground makes no sense. It must be replaced by a series resistor (in series with the last C) in order to allow a gain of app. -29 (or slighly larger).
(By the way: The referenced internet link contains more similar errors. Don`t blindly trust Internet sources).
However, the circuit can oscillate based on another principle: The opamp works as an inverting "Differentiator" (counterpart to an integrator) with a - more or less - constant phase shift of -90 deg. Both C-R chains must produce +90 deg at the desired oscillation frequency. But it must be emphasized that an oscillator with R-C lowpass blocks (instead of C-R high pass elements) - together with an inverting integrator - has much better properties (noise!).
EDIT: The oscillation frequency is w=1/(RC*SQRT(3)) and the oscillation criterion is fulfilled for Rfeedback=12*R.
EDIT2: Here is the loop gain:
T(s)=s³R²RfC²/(1+s*4RC+s²*3R²C²) ; Rf=feedback resistor.
Setting s=jw you find the oscillation frequency for T(s)=real (which means: Denominator must be imaginary, because the numerator is imag.).
EDIT3: In general, there are 4 different types of phase-shift oscillators:
It has been prooved that the last two alternatives are best.