(Unfortunately) This statement is absolutely correct(*). However, ciml is referring to toolbar PR difference of 0.03, not 3.0.
- they have a bug in the PR calculation
- they completely changed the PR formular
That paper has nothing to do with the ranking algorithm we are talking about but is just about the PR calculation. (Using well-known block techniques to speed up the calculation compared to the simple Jacobian iteration scheme.)
Also, Google modified the original PR formula already years ago, while the standard formula is still used in this paper. Another point is that Kamvar et. al. always (i.e. for d<1) consider the calculation as determining eigen vectors. This is unnecessary complicated (and from a principle point of view not correct). While the calculation for d=1 corresponds to the determination of an eigen vector, it corresponds to solving a set of linear equations for d<1. They also get problems with dangling pages (dead ends) which wouldn't appear if they simply solve the equation system.