@llenzi is correct — dada2 should not be used on pre-joined reads, since joining disrupts the error profile that dada2 uses to predict sequencing errors. This is only an issue for dada2; deblur can operate on joined reads.
70% loss is not unusual, and dada2 may well yield similar results. You should examine the output file to determine where these reads are being lost. It could indicate that many of your reads are shorter than the truncation length you chose; are chimeric; or are non-target DNA.
Yes something does not sound right here. As far as I can tell, multithreading is used at most stages of the dada2 pipeline so your job status is not consistent with expectations. Can you give us more information about your server? (number of cores, operating system, RAM, etc?)