Two approaches; the first is essentially the same as Rahul's, but slightly tidier (in my opinion).
Without loss of generality, we can shrink the radii of all three spheres until the smallest becomes zero. This will not change the equidistant locus. Also, we may assume that the zero-radius sphere, $S_0$, is located at the origin. Call the other two spheres $S_p$ (with center at $\mathbf{p}$ and radius $a$) and $S_q$ (with center at $\mathbf{q}$ and radius $b$).
At points $\mathbf{x}$ that are equidistant from $S_0$ and $S_p$, we have
$$
\|\mathbf{x}\| = \|\mathbf{x} - \mathbf{p}\| - a
$$
After some algebra, this gives us
$$
2a\|\mathbf{x}\| = 2 \mathbf{p} \cdot \mathbf{x} - \|\mathbf{p}\|^2 + a^2
$$
Similarly, points that are equidistant from $S_0$ and $S_q$ satisfy
$$
2b\|\mathbf{x}\| = 2 \mathbf{q} \cdot \mathbf{x} - \|\mathbf{q}\|^2 + b^2
$$
Eliminating $\|\mathbf{x}\|$ from these last two equations, we get
$$
a( 2 \mathbf{q} \cdot \mathbf{x} - \|\mathbf{q}\|^2 + b^2 ) =
b( 2 \mathbf{p} \cdot \mathbf{x} - \|\mathbf{p}\|^2 + a^2 )
$$
and re-arranging gives
$$
2\mathbf{x} \cdot (a\mathbf{q} - b\mathbf{p}) =
a\|\mathbf{q}\|^2 - b\|\mathbf{p}\|^2 +a^2b - ab^2
$$
This is the equation of a plane whose normal is in the direction $a\mathbf{q} - b\mathbf{p}$.
A Different Approach
The locus is the path of sphere (of varying radius) which is moving so as to remain tangent to the given three spheres. The envelope of such a moving sphere is a Dupin cyclide. The centerline of a Dupin cyclide is a conic. Details in this report.
