Nearly reversible computation will eventually replace conventional computation since reversible computation will be more energy efficient than conventional computation. Furthermore, symmetric encryption and decryption are very well suited for reversible computing, so we should expect for symmetric encryption and decryption to eventually be computed on reversible hardware or software. We should expect for future block ciphers to be eventually evaluated based on their performance on reversible hardware and reversible software. In fact, the advent of reversible computation will likely cause people to retire AES in favor of a more reversible block cipher before people would want to retire AES based on security concerns.
Reversible computing ubiquitously uses a technique called uncomputation which amounts to running the computation in reverse in order to clean up all the garbage information produced by the computation. A block cipher that is designed for reversibility should run on reversible hardware or software without any need for any uncomputation except for possibly the key schedule. In other words, in a block cipher that is designed for reversibility, not only must the encryption and decryption functions be invertible, but all of the components that compose the encryption and decryption functions should also be invertible. The process of uncomputation takes computational resources that are not being spent creating confusion and diffusion but are instead being spent reducing the amount of confusion and diffusion. Feistel ciphers tend to require some uncomputation while substitution-permutation networks do not require uncomputation (the most important component of a Feistel cipher is not invertible), so substitution-permutation networks will be more suitable for reversible computation.
It is probably a good idea for researchers to investigate reversible block ciphers right now to best prepare for encryption using reversible computers.