I'm having some trouble understanding virtual work and displacement, especially a particular section of Goldstein. I'll use an example to explain my difficulty, but I realize this might be the product of misunderstanding of virtual displacement.
Consider a pendulum whose length $\ell(t)$ varies as a function of time. It has two degrees of freedom: one along the rod, and one along its angular position; we can describe this system with generalized/polar-plane coordinates $r$ and $\phi$ and thus the constraint is holonomic.
As it swings, both $\phi(t)$ and $\ell(t)$ change and thus the displacement $d\mathbf{r}$ is not perpendicular to the constraint force of the rod. However, the virtual displacement $\delta \mathbf{r}$ points along the direction of angular motion, i.e., perpendicular to the rod at all times. Therefore we can dissect the displacement as
$$ d\mathbf{r} = \delta\mathbf{r} + \Delta\mathbf{r},$$
where $\Delta \mathbf{r}$ is a displacement along the direction of the rod, such that $||\Delta \mathbf{r}|| = \Delta \ell$ in that instant of time ($dt = 0$). We saw this decomposition also in this physics.SE answer on virtual displacement.
Now in Goldstein (section 1.4, p. 17) we have the claim $$ \sum_{i} \mathbf{F}_i \cdot \delta\mathbf{r}_i = 0,$$ where $\mathbf{F}_i = \mathbf{F}_{a,i} + \mathbf{f}_i$ is the total force, $\mathbf{F}_{a,i}$ is the applied force, and $\mathbf{f}_i$ is the constraint force.
Thus we could expand our sum for every $i$th particle to see that
$$ \sum_i \left[ \left(\mathbf{F}_{a,i} \cdot \delta\mathbf{r}_i\right) + \left(\mathbf{f}_i \cdot \delta\mathbf{r}_i\right)\right] = 0.$$
And this is zero by Goldstein's claim that the system in equilibrium. It is easy for me to see that $\mathbf{f}_i \cdot \Delta\mathbf{r}_i$ is nonzero because those vectors are parallel, and likewise that $\mathbf{f}_i \cdot \delta\mathbf{r}_i = 0$ because these vectors are orthogonal.
However, it is not clear to me (at least geometrically) why the other scalar product involving $\mathbf{F}_{a,i}$ is zero. Is this just a glorified statement of energy conservation? How does the original sum (a single term in my example) evaluate to zero for the pendulum?