Shareable nodes vs weak references

[Clemens Wyss]

Given the following data structure of projects and tasks:
project
    p1
        tasks
            all
                t1
                t2
            open
                t1
            closed
                t2

and I then want to add the employees/task association (tasks assigend to 
employees):

employees
    e1
        tasks
            open
                t1
            closed
                t2

The task nodes (t1 and t2) should, of course, not be replicated. Basically I 
see two options (of 5, 
http://osdir.com/ml/users.jackrabbit.apache.org/2009-10/msg00238.html ):
a) shareable nodes
b) multivalued weakreference properties

Which of these is better/more performant? Pros and cons of the two approaches? 
I would opt for a)...

Thanks in advance for your advices