In the paper "On Interprocess Communication", the author Leslie Lamport have developed a formalism for interprocess communication via shared registers based on lower-level, non-atomic operations and have proposed the definitions of safe, regular, and atomic registers. Also Lamport's description for the "interprocess" publication.
For safe registers, a read that overlaps a write operation is allowed to return any one of the possible values of the register.
My problem is
How practically relevant is this specification of safe registers?
Particularly, is it possible for a read of an actual register/memory unit in a multiprocessor computer to return an arbitrary value when it is concurrent with a write?
What about going to a lower level such as circuits/flip-flop?
In what sense, does a safe register exist?