For what it’s worth, I agree with Robbie’s comment that we should be cautious about drawing ordinal rankings from the RAE data. 

The point I want to make here is the obvious one that some of that information will be more useful for certain purposes than others, although all of those purposes could aptly be described as “assessments of research quality”.

For instance, a student wondering where to go to get the most great discussions with the most great researchers may be uninterested in the difference between department A with a large amount of 4* output and nothing else and department B with a similar amount of 4* output and also a large amount of 1* output.  If so she should not take averaging scores too seriously. 
 
To university admins, however, the difference between A and B might be extremely important for determining things like how much research bang for their salary buck they are getting.  They should take averages seriously.

Various different kinds of ordinal information might be obtained from the RAE data which might be useful for different purposes (though like almost everyone else I don’t think the methods used to formulate it were perfect), but I don’t think we should assume any ordinal ranking is best (or even approximately best) for all (or even most) research-quality-assessment purposes. 

UPDATE: I say this as someone who claims affiliation of one kind or another with a number of British departments, and is feeling neither particularly thrilled or nor particularly dispirited about any of their performances!