Memory locations (which is what the stringified reference reveals) are okay as long as you only care about them being unique for one run of one program in one process. For anything else, stay away. For example, I've seen people use memory locations as random session IDs for web code, which is not safe, especially in a cluster.

Personally, I don't think it's such a good idea to replace clean code that works (from chip) with code that is confusing for newbies and might not work if something changes about Perl's memory allocation scheme.