It's a deterministic process, but it still has inputs. You'd have to know those inputs or test for every possible set of inputs to whatever blurring method was used.
It depends on the blurring method. If you do a gaussian blur, for example, you might be able to do a pretty good inverse blur by finding black frames and performing blind deconvolution. If I were at risk I'd use a nonlinear injective filter instead, to make sure information is lost (linear filtering is not only deterministic but perfectly without noise and no frequency response nulls).
Given a known frame, they could blur the same portion of the same frame on all watermarked copies, and the one with a matching output is your culprit.
Blurring loses information, but is more akin to a hash than a deletion, as it's a deterministic process.
Edit: this sort of thing. http://dheera.net/projects/blur