UPDATE, June 26, 2009: Anyone interested in how I ended up developing this argument should check out the paper just published in PPR.
One of the more difficult issues for Kantian moral theorists is how, if at all, our moral obligations should be sensitive to others' wrongdoing. It seems fairly obvious that what we are morally required to do can change in response to others' immoral conduct. A clear example is promise keeping: If A and B agree to a mutually beneficial promise, but A doesn't fulfill the terms of their promise, B is presumably not obligated to fulfill them either. So A's wrongdoing influences B's moral obligations. Another example is punishment: Since punishment is the infliction of harm, suffering, or deprivation (which is typically wrong), it must be the case that the wrongdoer's wrongdoing justifies inflicting otherwise wrongful harm, suffering, or deprivation on her. This issue is acute for Kantians because Kantianism has long been seen as somehow more "principled" than consequentialism. The challenge for Kantians is to offer an explanation of how our moral obligations should be sensitive to others' wrongdoing that invokes key Kantian values or principles (rational autonomy, the categorical imperative, etc.) without becoming so sensitive to others' wrongdoing that Kantianism becomes indistinguishable from outright consequentialism.
The example that has of course stimulated much of the discussion surrounding this problem is Kant's treatment of 'the murderer at the door' in his essay "On a Supposed Right to Lie." There Kant seems to say that lying to a would-be murderer about the whereabouts of the innocent victim he intends to kill would be morally wrong. Most Kantians (and most reasonable people in general) find this conclusion troubling if not absurd: If ever there were a situation in which lying is not only morally permitted, but even morally required, that would be it!
What follows is my own (admittedly long-ish) attempt to answer the 'murderer at the door' problem in Kantian terms. Whether my attempt is of value in addressing the larger theoretical problem of how our moral obligations should be sensitive to others' wrongdoing, I'm not sure, but here goes: