I suspect with modern CPU pipelining and branch prediction that most of the (very interesting) debates about exactly how to tag values and pointers have become inconsequential above the noise level.

Would love to see recent work demonstrating this isn’t true!