This graph came up in our (Chekuri, Khanna and myself) search for constant integrality gaps for edge disjoint path problems in planar graphs. We had wondered whether such constant factor approximations were possible via well-linked decompositions: a partitioning of nodes into fractionally $\Omega(1)$-linked (or, informally, "well-linked") sets. The dual of this graph showed that a stronger form of this statement is false. A constant gap is ultimately obtained by decomposing into Okamura-Seymour instances.

The class of graphs can be thought of as being obtained as follows. Start with a circle $C$ whose (abstract, i.e., non-euclidean) length is $1$, and pick two antipodal points and join them by an edge of (abstract) length $1/2$. This is the graph $\Theta_1$ in the class. For each $i \geq 1$, $\Theta_i$ is a planar graph with $2^{i}$ faces each of whose boundary has an "outer" part, $O$, consisting of a segment of $C$, and an inner part, $I$, which is a curve inside $C$. Find the midpoints, with respect to their lengths, of $O$ and $I$ respectively and join them by a new edge of length $\frac{1}{2^{i+1}}$. Doing this to each face inside $C$ produces $\Theta_{i+1}$.

The graph shown is $\Theta_{11}$ and was drawn using python to produce script in metapost (John Hobby).