typical parameters for filtering Illumina RNASEQ reads

[PFS]

I've used fastq_quality_filter with q = 3 and p = 90 (at least 90% of each read must have quality >3). I thought these were relaxed settings and was expected only few reads being discarded (precisely those with all Bs). To my suprise, 90% of the reads were discarded when imposing the constraint that both mates must pass the filter.

Is that typical? Or perhaps I should let the aligner take care of the bad quality bases and keep all the reads in the dataset?

Thanks!

[rskr]

How long are the reads, and what is the median quality at each base?