Increase number of reads during subset step

Alignment are currently done on less than 0.1X for each run. It is not enough !