Dear Frank,
Your question doesn't have a simple answer, and relates to an active area of research. It's certainly not an uncommon finding to see negative hrfs that differ in magnitude between conditions. As for "relative to what," you can simply say that when the stimulus occurs, BOLD signal in that area tends to decrease in the following several seconds. It seems that some cases of this are indeed a blood-steal phenomenon, but not all. The chain of events leading from stimulus-induced neuronal input to negative BOLD deflections is poorly understood, but hey, the same is true of positive deflections. See papers by Raichle, Otten & Rugg, McKiernan KA, among others, for thoughts on the functional role of these deactivations. Opinions differ.
I'm actually in the process right now of writing a paper on the subject, relating fMRI deactivation to aspects of neuronal synchronization detectable in EEG. If you don't mind, I'd love to chat with you in greater detail than would be of interest to the AFNI forum - what your task is, what the areas are that show this effect, etc. You can believe that it is a real effect! Maybe we are both on to something similar. My email is jedmeltzerATyahooDOTcom.
-Jed