Yes I think the read-trhougth ebvent is you major issue here. The latest cutadapt log you posted looks so much better!
I wonder if a solution could be to remove primers in two stages as suggested in the following post:
(please look at the post marked as solution ).
Basically, they fisrt remove the primer in the tail of the sequences using the anywhere flag, then they trim the primer at th ebeginning of the sequences.
Hop it helps
Luca