Waves in spaces of even dimension
The phenomenon behind this is usually referred to as Huygen's principle, and can be described a little bit more precisely as the following:
Say you have the solution $u(x,t)$ to an initial-value problem for the wave equation in $x\in\mathbb{R}^n$:
$$\Delta u - c^2 u_{tt} = 0,\quad u(x,0) = \phi,\quad u_t (x,0) = \psi$$
It turns out $u(x,t)$ depends only on the values of $\phi$ and $\psi$ on the surface of a ball $B$, centered at $x$ and of radius $ct$, for odd dimensions $n \geq 3$. Heuristically, that means the initial "disturbance" always propagates out at $c$, so the "size" of the nonzero part of the solution (support) is invariant with time. For even dimensions, the value of $u(x,t)$ is also dependent on everything inside that ball $B$ (i.e. things don't just propagate at $c$), so the "size" of the non-zero part of the solution (support) isn't invariant with time.
A short explanation of why it happens is as a result of the fact that there is no transformation in even dimensions that can turn a wave equation into a 1-D problem, which is the case for odd dimensions $n \geq 3$. It is more thoroughly described and fully derived in Evans' book or here.
A slightly more beautiful but more esoteric explanation comes from Balasz, who noted that one can obtain solutions to the wave equation by contour integration after performing a Wick rotation. The integrands of these possess a differing type of singularity based on the dimensional parity; a branch point in even dimensions and a pole in odd dimensions. And when calculating the value of the contour integral in the odd $n$ case, one needs to refer only to one time value (that of the pole), while this is impossible in the even $n$ case.
I should also point out that a system of waves in even dimensions actually does asymptotically decay to nothing everywhere inside of the ball $B$ (i.e. everything in the interior of $B$ eventually destructively interferes with itself), as shown in this paper by Bob Strichartz; he notes this can be used to justify why the classical prediction of certain observables emerges out of free quantum wave-functions. So take Jim Holt's statement with a grain of salt!
The mathematics is explained here. It discusses the behavior of the Green's function for the wave operator in various dimensions.