Perhaps there is something here that'll help? Note, the comments about --p-match-adapter-wildcards probably do not help, as I think we set that as default in the later versions of 
.
Have you compared the trimmed sequence that still contains the primer with the original sequence? Perhaps there was a library preparation issue and/or sequencing error that resulted in multiple copies of the primer sequence? In which case you can set the --p-times 2.
Sometimes, there are a few almost identical sub-sequences within 16S that can make it appear that a primer is still present, when it is just a neighboring, yet very similar sub-sequence that is not actually the primer. Can you paste an example of a sequence (with the primer) before and after primer trimming, along with the primer sequence? I'd help to see the actual commands you're running too. 
-Cheers!