Hi Joseph,

You should NEVER look at point estimates for guidance, because they give you no indication of variance. For example, your site with dN = 6 and dS = 0 could have only one non-synonymous substitution along a short branch, and the LRT (which accounts for variance) is not significant, e.g. because dN = 6 +/- 3 and dS = 0 +/- 5 (hypothetically speaking).

The number of sequences is only one component of the "effective" sample size: FUBAR and FEL both derive signal from substitutions, i.e. if one site has more substitutions (roughly speaking), it will have a more precise estimate of dN and dS.

Sergei