One of the most compelling reasons for using amino acid sequences instead of
DNA sequences for drawing trees is convergences of codon usage. The third
position of most codons is redundant for all intensive purposes. A mutation
at this position does not change the encoded amino acid (note I said MOST).
In general mutations accumulate at this position with a higher frequency than other positions. It is possible that two organisms/genes that are not so closely
related and have independently *acquired* the same codon preferences will
show a greater degree of similarity than they should. The bias in preference
at the third position (and to a lesser extent the first position) is masked
by using amino acid sequences.
Hope I'm right.. :-)