Each to his own taste. I have a couple of Norinco monster cases that I set up originally, and wound up condensing down into the Node 804. It's a highly limited solution in terms of number of drives, stops pretty solidly at ten. But my needs for number of disks stops at seven. No need for the monster's huge number of drive bays, and the Node is quieter. I'm good at 12TB for the moment, even with room for expansion. Others may have higher storage needs.
I read Salter's article a ways back. He makes some good points. Probably the best points are that you can increase your pool in steps of only two disks of any give size, not many. IIRC, the thing he glossed over a bit was the crux of the issue. He was emphatic about being able to rebuild a replacement disk in a mirror "really fast".
I'd have been a lot happier if he'd quantified "really fast". :) The critical issue of course is whether you'll get enough disk failures in the vdev to lose the data. For a mirror, this is easy; you estimate the probability of a disk failing over any time period. If one disk has failed, your chance of losing the entire Vdev is the probability of losing that many more disks. The article properly puts this at the probability of losing one (the only remaining!) disk in the time remaining to resilver a new disk. For a mirror, you're betting your data that the resilver time is shorter than the time to the next disk failing.
For a RAIDZx, the probability is the probability of losing a second disk before the first loss can resilver, times the probability of a second loss given that one disk has already failed for RAIDZ1. The resilver time is indeed longer for a RAIDZ1 than a mirror, and there are a number of disks that can fail, so the probability of failure is larger by a factor of both the number of disks and the longer resilver time. For RAIDZ2, that's the probability of a THIRD disk going down times the probability that two have already died, in the resilver time. For RAIDZ3, it's the probability of a FOURTH disk dying during one resilver time for the array, given that three disks must have already died. I would have welcomed a clearer statement of the probabilities.
All parts can have a simplified estimate of the probability of failure in a given time frame. Disks, by and large, have a quite low probability of failure in any given small time interval. The bigger the time interval, the bigger the probability, and vice versa. Also some realistic estimates backed up with real-world data on resilver times.
I haven't done the calculations (obviously), or I wouldn't be winging this. But my gut tells me that it's better to bet that three or four disks won't die on the same day, let alone the same "resilver time", than to bet that a second disk won't fail while I'm refilling a new one. I may have to go get educated on the topic and run my own calcs.
Meanwhile, I'm waiting on a replication of my RAIDZ3 NAS to a purely-backup RAIDZ2 NAS backup. Once the replication finishes, I'll have both a RAIDZ3 front end, and a RAIDZ2 "mirror" of that. :) What are the odds??