The issue with crypto-shredding in many applications is that it aggressively pessimizes the software architectures possible in support of the data model. For things like databases and analytics, that feature increases the required hardware footprint -- compute and storage -- by orders of magnitude. The loss of efficiency has little to do with the crypto per se, since hardware support makes that cheap, but with how the change in representation interacts with idiomatic data structures and algorithms for high-efficiency data systems.
Many system internals are essentially a bag of tricks for constructing the maximally compressive representation of the important relationships in a data model while preserving practical tractability of operations over the data model. Crypto-shredding architectures have the effect of "encrypting" before "compressing", rendering the latter moot.