p-value different when compare two sample or three samples with cuffdiff

[safa]

I am have three different samples control (q1) and two treatments (q2 and q3). I am comparing the two treatment samples with cuffdiff and it generates in the gene_exp.diff file p-value that is significant while when I compare the three samples together, the gene_exp.diff file generates comparison between q1 versus q2, q1 versus q3 then q2 versus q3 and the p-value in this case between q2 and q3 becomes not significant. Why is that happens while the FPKM for the two treatment samples are the same? and which p-value I should use?

I appreciate your help.