What are some mechanics examples with a globally non-generic symplecic structure? In the framework of statistical mechanics, in books and lectures when the fundamentals are stated, i.e. phase space, Hamiltons equation, the density etc., phase space seems usually be assumed to be $\mathbb R^{2n}$, where maybe the $q^i$-coordinates are cut off to get a finite volume. 
In the books about Hamiltonian mechanics, especially mathematical books, one needs a symplectic space $(\mathcal{M},\omega)$ and of course the Hamiltonian. Now necessarily, locally $\omega$ looks like the canonical form $\Theta=\text dq^i\wedge\text dp_i$. 

Are there some relevant classical mechanics problems where one can state a less trivial $\omega$, and that globally? 

I would like to see a global expression which is different from $\Theta$ (and also not just $\Theta$ in different global coordinates).
That would be a nontrivial form, which might maybe arise over a more topologically complicated space than $\mathbb R^{2n}$, maybe due to restrictions of a mechanical system.  And maybe you get such a form after a phase space reduction, but I don't actually know any explicit mechanical problem you need it for. 
 A: Generally, coadjoint orbits of a Lie group provide important examples of global symplectic manifolds. In general such systems are obtained by symplectic reduction from a more fundasmental description.
For example, the spinning top is modelled for constant $J^2$ on a symplectic manifold $S^2$ that is a coadjoint orbit of the rotation group $SO(3)$. It is obtained by symplectic reduction from the $N$-particle model of a rigid body.
(If $J^2$ is not taken fixed, one needs a more general description in terms of a 3-dimensional Poisson manifold.)
There are lots of more advanced such models. See the book Mechanics and Symmetry by Marsden and Ratiu.
More generally, Hamiltonian dynamics in Poisson algebras is also not just a mathematical game but is important in applications. For example, the Hamiltonian description of realistic fluids needs an infinite dimensional Poisson manifold. For the Euler equation, see, e.g., 
P.J. Morrison, Hamiltonian description of the ideal fluid,
Reviews of Modern Physics 70 (1998), 467.
http://www.ph.utexas.edu/~morrison/98RMP_morrison.pdf
A: Phase spaces which are not cotangent bundles can be realized in
mechanical systems with phase space constraints .  The phase space given by
Arnold: the two sphere $S^2$ can be mechanically realized as the reduced dynamics
of an energy hypersurface of a two dimensional isotropic harmonic
oscillator:
$ |p_1^2|+|p_2^2|+|q_1^2|+|q_2^2| = E$
We observe that the Hamiltonian generates a constant rotation
rate in the $(p,q)$ planes, namely:
$ (p_i(t)+iq_i(t))  = exp(-iE_it) (p_i(0)+iq_i(0)) $
Thus we may choose to look at the system from the point of view of a
"rotating system in phase space" in which the vector in the $(p_1, q_1)$ plane is always
in the direction of $q_1$. Of course, we cannot do that on both planes
because we have only one degree of freedom. Thus we are left with:
$|p_2^2|+|q_1^2|+|q_2^2| = E$, 
which is just the equation of a two-sphere. Thus the reduced dynamics of
a constant energy hypersurface is on a two-sphere.
The symplectic form has to be  proportional the area of the sphere,
because it is the volume form of the sphere and a two sphere has only one
volume form.
This approach gives us a very big bonus upon quantization.
It is well known that from the quantization of a sphere we get spin
quantization. From the point of view of the isotropic oscillator for $E = 2j \hbar$, ($j$ is half integral), this
quantization corresponds to the following energies of the individual
oscillators: $(2j, 0), (2j-1, 1), .,.,., (0,2j)$. As can be seen there are
exactly (2j+1) states as in the spin system. 
The full theory of quantizations allows to write the corresponding wave functions also in the coordinates of the two sphere. Thus, we actually quantized the isotropic
oscillator using spin quantization. 
The equivalence of this procedure to the standard quantization of the isotropic harmonic
oscillator is a very celebrated theorem  by  Guillemin and
Sternberg called  "Quantization commutes with reduction". Actually, this
is the principle we apply when we quantize gauge theories (although there is no formal proof for the infinite dimensional case). You can find
on the net numerous works on this subject.
A: Any two dimensional closed orientable surface can have structure of a symplectic manifold (you can set your symplectic form equal to volume form). Moreover it will be "nontrivial" in the sense of being different from cotangent bundle of some other manifold. Also once you are given with some symplectic manifold you can always define a classical mechanical system on it, by introducing a Hamiltonian function and writing corresponding time evolution equations.
One explicit example is torus which can be obtained from phase space $R^2$ of a single particle by making following identifications on position and momentum : 
$x+L_1=x$ 
$p+L_2=p$ 
So now any function $H(x,p)$ which is periodic in $x$ and $p$ with periods $L_1$ and $L_2$ respectively can serve as a Hamiltonian function on torus.
