Some issues about DE genes detection by Cuffdiff

[wisense]

Hi,guys
I have some questions in the DE genes detection by using Cuffdiff.
I have 3 samples of paired-end RNA-seq data(Ct,Is,RP, Ct is an control sample,while Is and Rp are two treated samples), when I use Cuffdiff to find the DE genes among these 3 samples by just using all the tophat results of the 3 samples in an Cuffdiff run, the DE genes results are a bit different from that of using two samples in an Cuffdiff run. The command lines are as follows:

Use all the 3 samples:

cuffdiff -o diffout_all -L Ct,Is,Rp -p 12 -b mm9.fa mm9.gtf Ct_tophat.bam Is_tophat.bam Rp_tophat.bam > cuffdiff_all.log

Use 2 samples in one Cuffdiff run and run Cuffdiff 3 times:

cuffdiff -o diffout_CtvsIs -L Ct,Is -p 12 -b mm9.fa mm9.gtf Ct_tophat.bam Is_tophat.bam > cuffdiff_CtvsIs.log 
cuffdiff -o diffout_CtvsRp -L Ct,Rp -p 12 -b mm9.fa mm9.gtf Ct_tophat.bam Rp_tophat.bam > cuffdiff_CtvsRp.log
cuffdiff -o diffout_IsvsRp -L Is,Rp -p 12 -b mm9.fa mm9.gtf Is_tophat.bam Rp_tophat.bam > cuffdiff_IsvsRp.log

Can anyone tell me how does this difference happen? Which way is correct to do DE genes detection when you have this type of data?

Thanks very much~
wisense