Recommended use for PE reads with varying insert sizes?

[cladobranch]

Dear all,

I would like to use RSEM-EVAL to evaluate the quality of several denovo transcriptome assemblies from Illumina PE reads. The sequencing libraries were prepared with a varying insert size, which is why I have been wondering if you would recommend

1. using RSEM-EVAL with the option --paired-reads indicating the average insert size of the library OR
2. running RSEM-EVAL as though the library was made up of single reads indicating the average read length.

Could you tell me which of the two is more likely to produce an accurate evaluation of my transcriptomes?

Cheers, Dario