https://en.wikipedia.org/wiki/Cyclic_redundancy_check#Standa... suggests that Castagnoli has better "error detection capacity". What that means exactly (in terms of information theory) and whether that difference matters in practice is getting beyond my expertise level. In any case, existing formats like Zip are what they are, and as I mentioned in the blog post comments, it is still possible to SIMD-accelerate the IEEE flavor.