I suggested this on the initial AF447 thread, and didn't get much response, but I don't see why it would not work:
As long as you have a valid IAS, you systematically compare it to GS using GPS and compute wind speed and direction from that. You store this value. If the IAS fails, you use this stored value to correct the GPS-derived GS to get an estimated AS.
Except in the case of severe wind shear, why would this not work?
You could present the estimated AS as a range, with some degree of uncertainty, incorporating previous variations in wind speed, if any, or just an uncertainty factor (perhaps growing over time) to account for possible changes in wind speed.