There's no way at all in the general case (a function name could be passed as a parameter, for example). I think Matthew is suggesting to track dependencies at run time, but that seems a recipe for burnt fingers and an overall performance loss, for what seems a dubious gain.
I wouldn't think that it would be that much of a performance loss (except for DDL type things where you were setting up the dependency in the first place) though you would certainly know better than me. In any case, I know it would be a lot of work and I wasn't suggesting doing it all at once, but rather as a rule of thumb going forward - a continuous improvement to be made as other work is done. If people in a better position than I to gage this really think performance would suffer a great deal or that it's not doable (or shouldn't be done), I respect that. It seems that we should at least give warnings where we can though and probably doc it somewhere. "hey user, you replaced a plpgsql function that was used in an index, your index may now be invalid. Please see deatils at xyz".
