The clawback plugin improves species-level classification, but this is meant for realistic datasets so may not help in your case with mock communities (you could make taxonomic weights based on the expected composition of that community, but that would be overfitting to that mock community and not perform well for real samples):
QIIME 2 is just wrapping these other methods, so I suggest that you read the documentation for vsearch, dada2, and deblur to understand how their chimera checking methods differ.
No trimming means the sequences will contain more errors, meaning that dada2 is more likely to throw out those sequences during the pre-filtering step. You should always check out the denoising stats artifact to determine how many reads are being lost at each step.
I hope that helps!