If I understand correctly, you are sequencing transcripts? It that is the case, it is quite possible that the orphaned reads are due to the lack of alignment across intron-exon boundaries. If I recall, Mosaik is not designed to align RNA-seq reads. Have you considered using an RNA-seq aligner such as GSNAP or tophat?

See the answer by Sean Davis if you are mapping exons on a genomic reference.

For finding large deletions and insertions, or other types of translocations, you could try Pindel. Its pattern growth algorithm does exactly what you are looking for.

Coincidentally, we found that GSNAP also works fine on DNA to find large deletions.

A common problem with gene expression studies and sequencing - ESTs, RNA-Seq, etc - is contaminating genomic DNA. Some mRNA preps are excellent and some are poor. There is always some amount of genomic DNA, or unspliced or incompletely spliced messages in the mix. These could be a (partial) source of the orphaned reads. So, you can see if any orphans align to a contiguous segment of the genome and if so, if any of that alignment falls within intron. In some cases a retained intron is a legitimate splice variant, but this is rare, and would not be expected without first seeing matches to the known gene models. Thus, too many orphans mapping to introns is likely a sign of issues with the mRNA prep.