We made a conscious decision NOT to override the default 
implementations of equals and hashCode, and rely exclusively on object 
identity. There are the known limitations, of course, but they haven't 
been much of a problem in practice. We rarely have a need to "mix" two 
objects with the same database identity loaded from two different 
sessions. And when we do, we create an inner class that encapsulates 
the business key from the entity and use that as the key into maps and 
such.

It may not address all cases perfectly, but it's consistent and easy 
to follow.