BLAST only ever works with single queries so will report data for each one separately. If you want a single, 'overall' match percentage you can determine it by taking the average of all the single queries.

However, a single overall measure is rarely very meaningful. A boxplot or histogram describing the distribution of transcript similarities will be a nicer way to represent the information.

You could use cd-hit-est-2d to cluster the two transcriptomes at several thresholds, starting from higher similarities and then decreasing.

edit: BLAST is local search algorithm, and it is not suitable for estimating match percentage, as it will generally inflate it as it leaves out from the alignment portions of the transcript that not pass a certain threshold score. See here for a discussion on this (on another context, though).

Try Mauve. It is a nice program to compare genomes and transcriptomes

I think you need to rethink your question. You have transcriptimes from non-model species. That means that you only have a rough idea of the gene number in these organisms. What I suggest is a comparative approach. Why use blastn? You are comparing transcriptomes here, coding sequences mostly. Try tblastx of species 1 vs species 2, sequences that return no hits are unique to species 1. Then do the same thing for species 2 vs species 1, and find which transcripts are unique to species 2. Then blast the unique transcripts with blastx vs nr to find out what is unique to each species.