After making some tests with support team, update drivers and firmware, etc, it seemed my job was working well, and amount of tape usage was corresponding to backup usage on disk. I deleted and recreated my weekly backup to delete any previous weird data from older updates or something...
Well, the first iteration worked well, but one week later, on the second iteration, my job processed 1,6TB of data and occuped 2TB on tape again. My job is only copying the last full backup always (reverse incremental backups and no incremental processing on the tape job), so is not an incremental or synthetic issue.
I'll try this week to delete manually the tapes before the tape job starts, to see if it makes any difference, but it is an annoying problem.
I've also set up the block size to 256k manually (default for my tapes is 128k).