Why is the parity of the spatial wavefunction $(-1)^{\ell}$? Consider a composite particle state $|\psi\rangle$ (like a hadron or a meson) that is an eigenstate of some Hamiltonian (e.g. the QCD Hamiltonian). Since the Hamiltonian is invariant under rotations and parity this particle state is also an eigenstate of the angular momentum and parity operator:
$$L^2 |\psi\rangle = l(l+1)|\psi\rangle$$
$$P |\psi\rangle = (-1)^{a}|\psi\rangle$$
where $a$ is an integer number. Why is $a = l$?
For two particles one can use the 'trick' to transform into relative coordinates and then find that in relative coordinates the eigenfunction is $\sim Y_{lm}$. The parity of the spherical harmonics then leads to $(-1)^l$. 
I don't see how to extend this to 3 or more particles.
EDIT:
I had the following idea how to extend to 3 particles:
For 3 particles the Hamiltonian looks like: 
$$H = \frac{p_1^2}{2m_1} + \frac{p_2^2}{2m_2} + \frac{p_3^2}{2m_3} + V_1(|x_2-x_1|) + V_2(|x_3-x_1|) + V_3(|x_3-x_2|)$$.
Now choose new coordinates by
$$R = \frac{m_1 x_1 + m_2 x_2 + m_3 x_3}{M}$$
$$y = x_2 - x_1$$
$$z = x_3 - x_1$$
The Hamiltonian becomes:
$$H = \frac{p_R^2}{2M} + \frac{p_y^2}{2\mu_{12}} + \frac{p_z^2}{2\mu_{13}} + \frac{p_y\cdot p_z}{m_1} + V_1(|y|) + V_2(|z|) + V_3(|z-y|)$$
where $\frac{1}{\mu_{ij}} = \frac{1}{m_i} + \frac{1}{m_j}$ are reduced masses
The total angular momentum is given by
$$L = x_1 \times p_1 + x_2 \times p_2 + x_3 \times p_3 = R \times p_R + y \times p_y + z \times p_z$$.
The $l$ in parity $ = (-1)^l$ is given by the internal angular momentum
$$L_i = y \times p_y + z \times p_z$$ which commutes with the Hamiltonian.
Therefore an eigenfunction is given by $$|\psi(y,z)\rangle = |f(|y|,|z|)\rangle |L M\rangle_{\hat{y}\hat{z}}$$. Using Clebsch-Gordan coefficients this angular can be written as:
$$|L M\rangle_{\hat{y}\hat{z}} = \sum_{m m'} \langle lm,l'm'|LM\rangle Y_{lm}(\hat{y}) Y_{l'm'}(\hat{z})$$ for some $l$ and $l'$
The overall parity is given by $(-1)^{l + l'}$ which does not necessary equals $(-1)^L$. For example $l = l' = 1$ would lead to a (from my point of view valid) solution:
$$|10\rangle_{\hat{y}\hat{z}} = \frac{1}{\sqrt{2}} \left(Y_{11}(\hat{y})Y_{1-1}(\hat{z}) - Y_{1-1}(\hat{y})Y_{11}(\hat{z})\right)$$
with parity $(-1)^{l+l'} = (-1)^{1+1} = 1 \neq (-1)^1 = (-1)^L$.
There must be something which excludes such combinations. Why is this solution not valid?
 A: $\def\br{\mathbf r} \def\bR{\mathbf R} \def\bl{\mathbf l} 
\def\bL{\mathbf L}$ 

Consider a composite particle state $|\psi\rangle$ (like a hadron or a
  meson)

I don't see where did you take into account particle's compositeness,
nor you had to. Actually this a general QM matter - no need to bring
up QCD or the like.
For one particle in a fixed potential your argument of spherical
harmonics applies. For two particles interacting with each other but
otherwise free, the same argument applies to relative coordinates. 
For three particles (or more) you follow the same route, only with a
somewhat higher complication. Choose (judiciously) two of the
particles, introduce their c.o.m. G, then c.o.m. of G and third
particle. Thus you have two position vectors: $\br$, going from
particle 1 to particle 2, and $\bR$, going from G to particle 3.
It can be shown that kinetic energy splits into two terms, one
depending only on $\br$ and the other on $\bR$. Then you may choose a
basis of eigenfunctions of two angular momenta, say $\bl^2, l_z$ and
$\bL^2, L_z$. You see that total wavefunction has parity $(-1)^{l+L}$. This is rather tricky, as one could erroneously be drawn to believe parity depends on total angular momentum, whereas it's not so: it depends on the sum of separate quantum numbers.
I talked only of kinetic energy, but of course in order parity may be a useful quantum number potential energy (or interaction lagrangian in QFT) are to be invariant under space reflections.
A: $\def\bL{\mathbf L} \def\bl{\mathbf l} \def\bp{\mathbf p} 
\def\br{\mathbf r} \def\bP{\mathbf P} \def\bR{\mathbf R} 
\def\frac#1#2{{\textstyle {#1 \over #2}}} \def\half{\frac12}$
In order to reply your objections it's best for me to write some
equations. As a general comment: you should not think of a change of
reference frame, but only of expressing the original quantities
(hamiltonian, angular momentum) in terms of new coordinates. Let's see
how.
I'm going to assume all mases are equal, just to make formulas
simpler. But you may verify that the argument is quite general. On the
other hand, it's a time-honoured approach, known as Jacobi's
coordinates and widely used in celestial mechnics more or less since
mid-19th century.
Call $\br_1$, $\br_2$, $\br_3$, the position vectors of the three
particles. Define
$$\bR = {\textstyle {1 \over 3}} (\br_1 + \br_2 + \br_3)$$
$$\br= \half (\br_1 + \br_2) - \br_3.$$
$$\br' = \br_2 - \br_1$$
Kinetic energy:
$$K = \half\,m \left(3\,\dot\bR^2 + \frac23 \dot\br^2 + 
      \half {\dot\br'}^2\right).$$
Conjugate momenta:
$$\bP = 3\,m\,\dot\bR \qquad \bp = \frac23 m\,\dot\br \qquad
  \bp = \half m\,\dot\br'$$
$$K = {P^2 \over 6\,m} + {3\,p^2 \over 2\,m} + {{p'}^2 \over m}.$$
Angular momentum:
$$\bL_{\mathrm{tot}} = 
  \bR \times \bP + \br \times \bp + \br' \times \bp' = 
  \bL + \bl + \bl'.$$
Commutators are those expected for canonical coordinateses $\bR$ with
$\bP$, $\br$ with $\bp$, $\br'$ with $\bp'$.
Parity quantum number refers to the transformation
$$\br \to - \br \qquad \br' \to -\br'$$
and consequently
$$\bp \to - \bp \qquad \bp' \to -\bp'.$$
Eigenstates of $\bl^2$, $\bl'^2$ have parity $(-1)^{l+l'}$ and I see
that you say the same thing, even if with different coordinates (which,
incidentally, gives rise to a non-separable hamiltonian).
What I cannot understand is why you are troubled with your example. You
built a state of internal angular momentum 1, $z$-component 0, starting
from states with $l=l'=1$, then parity +. What's wrong with that? It's
a perfectly possible situation.
