Vaughan, the equilibrium sensitivity is not used to project temperatures to 2100; for the reason you mention. We don’t expect 2100 to be in equilibrium.
To see the IPCC projections to 2100, refer to AR4 technical summary section 5. In all cases, the projections are simply based on models.
Some people, mostly non-experts I think who are misusing the concepts, may use sensitivity to project temperature in that way; but in the literature I think the major use of equilibrium sensitivity is as a diagnostic for comparing or characterizing models.
For example, Hansen’s work in 1988, which has been discussed here, is universally recognized, in Hansen’s own retrospectives also, as “running hot”. The model has a sensitivity that is rather higher than the more advanced models in use today. That’s an example of a straightforward use of sensitivity as a model diagnostic.
Cheers — Chris