Even with 67 markers Y-DNA STR testing is not a very fine tool for pinning down the number of generations. The 95% Confidence Interval is 1-10 generations. The 2.5% on the upper side has a very long tail. Very remote cousins are often better matches than very close cousins.
You can see it when you use Doug McDonald's TMRCA calculator to see the probability distribution, rather than the cumulative probability distribution. Note that he uses transmission events (meioses) which are twice the number of generations. (FTDNATiP and Doug use different mutation rates than I did in my calculation above.)