Consider the entanglement dilution protocol for pure states, as described in Preskill's notes (Link to pdf, see around page 32). The context is that Alice and Bob share $k=n(S(\rho_A)+\delta)$ Bell pairs, with $S(\rho_A)$ the von Neumann entropy of $\rho_A\equiv \operatorname{Tr}_2(|\psi\rangle\!\langle\psi|)$, while Alice holds $n$ copies of $|\psi\rangle$. Their goal is to end up, performing only LOCC operations, sharing $n$ copies of $|\psi\rangle$. They thus want to somehow "convert" the entangled pairs they share into another form of entanglement.
As described in the notes, the idea of the protocol is roughly as follows:
Notice that asymptotically in $n$, Alice's states have the form $$|\psi\rangle_{AC}^{\otimes n} \sim \sum_{\vec x}\sqrt{p(\vec x)}|\vec x\rangle_{A^n}\otimes |\vec x\rangle_{C^n},$$ where the sum is over typical sequences $\vec x$, that is, roughly speaking, sequences containing a number of 0s and 1s close to their expected value, so that also $p(\vec x)\sim 2^{-n S(\rho_A)}$. Here, $A$ and $C$ are two degrees of freedom held by Alice.
Alice teleports the $C^n$ half of (the projection onto the typical subspace of) $|\psi\rangle_{AC}^{\otimes n}$ to Bob, using only the $\sim n H(\rho_A)$ Bell pairs.
Bob performs a decoding operation on his register, and the result is them sharing $|\psi\rangle^{\otimes n}$.
I'm not quite sure how the teleportation step works here. In particular, why can Alice manage to teleport the $2^n$ qubits in the $C^n$ register, using only $\sim n H(\rho_A)$ Bell pairs?