Committed sumcheck in hiding and Translator (currently only in the ECCVM) has the potential to reduce proof size by a non-trivial amount. In each round of sumcheck we send a univariate of length equal to the highest relation degree + 1. If this is ~9 field elements and we can reduce to one commitment plus two evals (at 0 and 1), this is a 3x reduction (since commitments can be represented in only 32 bytes with point compression).
One downside I see is the additional openings would add a small number of ops to the ecc queue. Maybe there are other downsides/obstacles that I'm overlooking.
If everything looks good, maybe resolving this issue is just doing the implementation.