I'm not sure what "Pauli's argument" precisely is because he refers to pages in the first edition of Dirac's Principles of Quantum Mechanics which contain nothing of evident relevance in my fourth edition, but the more common thing to say is that it is the boundedness of the energy from below that forbids a naive time operator, not the discreteness.
However, the "particle on a ring" scenario you propose is not so simple: The space of wavefunctions on a circle may be naively written as $H = \{f\in L^2([0,2\pi],\mathrm{d}\phi)\mid f(0) = f(2\pi)\}$, i.e. all the possible square-integrable function with periodic boundary conditions. Multiplying by the "position operator" $\phi$, you must observe that $\phi$ does not preserve the boundary condition: $(\phi f)(0) = (\phi f)(2\pi)$ if and only if $f(0) = 0 = f(2\pi)$. So either you have the unnatural restriction to wavefunctions vanishing at a particular point, or you must relax the boundary conditions for a general wavefunction (none of these have direct physical meaning because the value at the single point does not change the probability density). For more on where exactly the commutation relation is valid, see this answer of mine.
The issue here is that the commutation relations themselves involve unbounded operators and are rather pathological. If they only hold on a subset of the space that is not the same as the dense domain on which the operators are defined, which can happen as in the case of the particle on the ring, then the Stone-von Neumann theorem does not hold and it is possible to construct all sorts of counterexamples.
The formally correct way to state Pauli's theorem is then this:
If the Hamiltonian $H$ is bounded from below or has a discrete part of the spectrum, then there is no operator $T$ such that the Weyl relations
$$ U(t)V(e) = \exp(-\mathrm{i}te)V(e)U(t)$$
for $U(t) = \exp(\mathrm{i}Ht),S(e) = \exp(\mathrm{i}Te)$ hold everywhere.
This follows directly from the Stone-von Neumann theorem, since unitarily equivalent operators necessarily have the same spectrum, and the canonical representation of the Weyl relations has both operators continuous and unbounded. It means that although you might be able to construct the infinitesimal version of the time operator in some cases on some part of the Hilbert space, you cannot get the "tame" version of the CCR, the Weyl relations, from them. To see that Weyl relations are really physically necessary for something to be the proper "time operator", note that $U(t)V(e)U(-t) = \exp(-\mathrm{i}te)V(e)$ is really just the statement that the time translation operator $U(-t)$ shifts the value of the time observable by $t$.
