There was a recent comment made by Gregory Maxwell regarding Confidential Transactions vs ZCash pruning and performance:
CT is currently faster to verify, though its pretty close in that metric. The rest of what you said holds. The size difference isn't quite so clear cut: not every CT transaction needs a proof (only ones that split coins do), and 80% of the CT proof size can be used to communicate private metadata to the far end.
CT+CJ (and CT+OWAS, which isn't well known yet) is also fundamentally pruning compatible while ZC depends on ever-growing accumulators to track spendable/spent coins. This, even more than the new cryptographic assumptions made by ZC, is why I think that the CT+CJ approach is interesting, important, and more practical-- even though it provides a smaller anonymity set.
I wonder how big a problem this is and if there is any workaround or incentive possible to minimize the long-term cost.