Superdense coding specifically is unlikely to ever be practically relevant. Even for people quite optimistic about the future of quantum computing it is hard to imagine a scenario where a single qubit will be cheaper than two classical bits.
Other quantum protocols are much more likely to be useful, perhaps particularly for cryptography.
I sometimes consider a qubit to be worth two classical bits in a very heuristic sense. One can superdense-code two bits into a qubit. Similarly, one can send two classical bits to teleport one qubit.
Both of these require spending entanglement, so it’s not a very useful heuristic.
The cost is two classical bits and one entangled Bell pair to teleport a qubit. You make the Bell pair, send one half of it to Alice (who wants to transmit the qubit) and one half to Bob (who wants to receive it). Alice then does a Bell basis measurement on the qubit she wants to transmit and the half of Bell pair she owns.
The Bell basis measurement has four outcomes. For convenience you can label them
00
01
10
11
Alice tells Bob the measurement outcome she got (this is the part that requires two bits of classical communication), and depending on which outcome she got Bob has to apply either no correction or a Pauli X, Y or Z gate.
But they are vague about the implications and I'm generally unsure about the practical relevance of quantum protocols like superdense coding ...