In general, I agree that prosecutors should refuse to go to court with a criminal case if the case hinges on testimony from a person who cannot be trusted. This is, in general, a sound civil libertarian stance. However, one need only read Radley Balko's blog, or the ACLU's blog, or any number of reports on death penalty cases with shaky evidence, to recognize that this standard of credibility should not only be applied to victims of alleged sexual assaults. It should also be applied to informants (especially in death penalty cases), cops who can't keep their stories straight, and "expert" witnesses offering testimony that is dubious at best.So, by all means, if (for the sake of argument) DSK's accuser is a person who cannot be trusted, and no other substantial evidence can be mustered to corroborate her allegation, drop the case. Better to let a hundred guilty people go free than send one innocent to prison. However, let's be just as scrupulous in examining the credibility of other prosecution witnesses as well. Credibility considerations shouldn't be exclusively for accusers in rape cases.
I think the issue for a lot of advocates of rape survivors is that you get prejudice passing for "credibility." The fact that you at various points in your life seemed to have enjoyed sex, for instance, isn't really about credibility.
With that said, I would not have credibility tossed from the process. I'd just have it fairly applied. I can, with little to no effort, show you men who are in jail on the testimony of witnesses whose credibility problems far out strip DSK's accuser. We are not talking about lying on tax returns, we are talking about racist drug addicts who send people to death row. We are talking about gang-affiliated cops who plant evidence.