Stealing bits from pointers makes sense sometimes, but in most cases an easier and less error-prone way to increase memory efficiency is to just use indices or relative pointers. Instead of shaving individual bits of a pointer take advantage of the fact that most pointers in a program are almost exactly alike.

Instead of storing a 64 bit prev and next pointer for an item in a linked list just store e.g. two 16 bit offsets. Then the next_ptr = (this_ptr & base_mask) + offset * alignment.

Most of the time wasting memory by storing full 64bit pointers is no big deal. But when it is, you can go a lot further than just shaving off a few bits.

If you want to be extra clever you can encode values in the base address of your pointers by taking advantage of how virtual memory is allocated. Every process has its own virtual memory pages and you can request memory pages at specific address ranges. For instance if you align all your memory at a 32gb boundary then you can use 32bit pointers + a base pointer without even having to mask. You can also use virtual memory allocation tricks to convey information about alignment, typing, whether a pointer is allowed to be shared between threads and so on. And again, without having to butcher the pointer itself.