Welcome to the forums! 
This is a great first post! I really like your approach of trying multiple joining settings and seeing what works best.
Ver 3 (Forward: 250; Reverse: 250)
This result is fascinating, as it shows that merging works best with the longest reads you have.
You may have found this already, but I bet the length of the V3-V4 region is causing problems.
How many basepairs long do you expect this amplicon to be?
With 250x2 bp reads, how much overlap do you expect with no trimming?
(here is how to calculate the length of overlap during merging)