Gordon Yorke (EclipseLink Architecture Committee Member, TopLink Core Technical Lead, JPA 2.0 Expert Group Member) wrote a good answer on this topic so instead of paraphrasing him, I'll quote his answer:
The difference between optional and
nullable is the scope at which they
are evaluated. The definition of
'optional' talks about property and
field values and suggests that this
feature should be evaluated within the
runtime. 'nullable' is only in
reference to database columns.
If an implementation chooses to
implement optional then those
properties should be evaluated in
memory by the Persistence Provider and
an exception raised before SQL is sent
to the database otherwise when using
'updatable=false' 'optional'
violations would never be reported.
So I tried the @Basic(optional=false) annotation using JPA 2.1 (EclipseLink) and it turns out the annotation is ignored in actual usage (at least for a String field). (e.g. entityManager.persist calls).
Basic(optional): Whether the value of the field or property may be
null. This is a hint and is disregarded for primitive types; it may be
used in schema generation.
So I think this sentence explains the real use case for Basic(optional) it is used in schema generation. (That is: when you generate CREATE TABLE SQL from Java Entity classes. This is something Hibernate can do for example.)