No problem. 
It will not hurt to throw it in, but you'll have to do the work of orienting the reads. That is, if these are indeed in mixed orientation. Your sequencing facility should have provided you with all of the details on how your data was generated.
When I see low values like this, it is often due to using the incorrect primer sequences. If you read through this post you'll see that I inquired as to which V3V4 primer was used. For example, your reverse primer is slightly different from two other commonly used sequences:
| Your 805R primer | Alternate 805R primer | 785R primer |
|---|---|---|
| GACTACNVGGGTWTCTAATCC | GACTACHVGGGTATCTAATCC | GACTACHVGGGTATCTAAKCC |
I suspect that the revers primer you are using is incorrect and might be one of the other two? 
Perhaps give one of those other ones a try, or add all of them to your cutadapt command.
They may be detected in the 3' end due to potential sequence read-through. Not because they exist in the 5' end. Though I would not expect this to be the case for the V3V4 region, only for very short amplicon data. But, if much of your data is less than the typical length of the V3V4 region (~460 bp), then this could be why), 
Unless you know for sure that your reads are in mixed orientation then, I'd stick with searching for primers in the correct orientation from the 5' end. Otherwise you'll risk spurious sequence trimming. Which may be what is occurring here.
Have you tried without using the reverse compliment sequence? It is weird that the "pairs too short" drastically increased. That is it could be spuriously trimming / finding a primer towards the 3' end of the sequence (the reverse compliment) and when it trims... it is removing everything from that primer sequence back towards the 5' end (i.e. the begining) of that read.
Perhaps manually inspect several of the sequences and see where these primers (and their reverse compliment) reside within your sequence.
See this thread about minimum read overlap required to merge reads.