in their manual they state:
'...the RNA-Seq fragment counts can be used as a measure of relative abundance of transcripts, and Cufflinks measures transcript abundances in Fragments Per Kilobase of exon per Million fragments mapped (FPKM), which is analagous to single-read "RPKM", originally proposed in:....'
so, I guess they just calucalte for each of the fragments (which is here chr1 and chr2) the FPKM, thats it. Since they don't incorporate the paired end info here....
imho, they don't have to as long as those reads map to a certain position in a unique fashion...
I have the same question and I went through their website. What I believe from their explanation is that they estimate fragment lengths for paired-end reads from the reads mapped to single-isoform genes. Given that they give the transcript abundances in terms of FPKM, where a fragment comes from two reads that are paired, I believe they would disregard improperly paired reads.
Also if you are using TopHat to align the reads, I think you can use "--no-discordant" option to remove reads where the mates are aligned to different chromosomes.