At ALSE, we’ve performed this kind of jobs a few times already.
In some cases, a lot of the original ASIC specification was lost (but we got high level description and a netlist) , at other times we had more complete functional description as well as expected values based on known stimuli.
These projects were not trivial but all successful in the end. In one case, a tweaking phase was necessary to painfully re-create the bugs and limitations of the original ASIC (!) since they contributed to create enormous databases of data that were exploited by sophisticated algorithms that took into account the ASIC exact behavior.
It is certainly our typical projects, but we’ve been there, done that ;-)