If, as it seems, you are ultimately binning the same data but sorting it on somewhat different metrics, it makes sense that your top numbers would agree, but having your lower numbers disagree by so much is disturbing.
Any idea what might be going on?

One factor: I always smooth my forward-looking return numbers.
So, for example, when looking at the one year return from today, I will compare today's exact price to the average price 11-13 months from now, or 10-14 months from now, so that the signal is primarily a function of the particular situation on the starting date, not the luck of the draw on the end date. This dampens extremes quite a bit. Especially the bad returns, since extreme low points tend to be short lived outliers while highs are more leisurely.

You wouldn't want to place much weight on the predictive power of P/B on March 5 2004 to estimate typical five year returns, because the end date would be March 5 2009, the bottom of the credit crunch drop. P/B dropped 48% in that particular interval...

Jim