Boot to Ghost, install os, play, run, do whatever. In the event of a virus, it's short lived. When I attended Berkeley, this was the way they had set up their computer lab. I remember, at the time, being intrigued by the setup.

I have no knowledge of Ghost (use dism & ntfsclone primarly) but doesn't the "Run" step require a reboot ?
Does your user know how to boot Ghost manually or is there a special bootloader that gets updated on deploy and when Windows is first ran.
Are your images Sysprepped or machine specific ? And if so aren't you bothered with the lengthy "pseudo-setup" phase (eg Windows welcome (OOBE), Installing drivers, ...) ?

The non-biological components are definitely more robust. You can just plug in new ones. Even do upgrades. The first thing I'd do after the amputation is dedicate a bunch of time to learning as much about prosthetics and 3D printing as I could. "Doctor heal thyself" would be a pretty good thing to dedicate the rest of my life to.

Bullshit, your biological components are a lot more robust. The average llifespan is 80 years.
Show me one mechanical machine (heart) which can do that without some offline maintenance.

Prosthetics are a long way off from "superlimbs".
It takes proper training to be able to even use current gen prosthetics.
And both scenario's are coupled with a lot of pain and hardship: nerve damage, ghost limb syndrome, ...

AMD has benched their "upcoming" ARM products. (http://arstechnica.com/information-technology/2015/04/amd-abandons-the-microserver-market-takes-75-million-hit/)

So poison the oceans, suffocate the atmosphere, irradiate the land because in a few decades I won't be around anyway.
Selfish religious assholes !

But you are equipped to put forward your prescient feeling of a mini ice age coming really really soon as something all of us should plan for ?

I wonder if it is possible to start a class action lawsuit ?

Yes the dreaded ST3000DM001-1CH166. Terrible that one. We've had 17 out of 24 fail in a consumer NAS fail in less than 3 years. It's also a very widely deployed external drive since it's cheaper than WD and WD has its own external drive products.

0 failures on the spinning disk, but I will give it a few years.

You haven't used any recent Seagates then :D. When I look to my right I see two spinners being recovered with ddrescue :(.
And it's been like that for the last 2 years. These large spinning consumer drives are just crap.

Have you tried different brands ?

As I posted above, SMART 177 isn't a percentage, it's a P/E count. Documentation claims 2000 P/E cycles, independent testing shows close to 6,000. It's far less than even 1% used.

You're absolutely right. I wonder what other SMART values will be really important to SSD's.
Where is blackblaze's SSD data. My guess is it will be even more vendor specific as it is now.

Well no.. You're at 8 write cycles used, and 99%.

Right that value represent the raw maximum erase cycles (~3.000 cycles TLC, ~10.000 MLC, ...)
Let me quote wikipedia:

173 0xAD SSD Wear Leveling Count Counts the maximum worst erase count on any block.

Now my own raid0 drives:

Device Model: Samsung SSD 840 EVO 750GB
Firmware Version: EXT0DB6Q
User Capacity: 750,156,374,016 bytes [750 GB]

9 Power_On_Hours 0x0032 098 098 000 Old_age Always - 5954
177 Wear_Leveling_Count 0x0013 099 099 000 Pre-fail Always - 6
241 Total_LBAs_Written 0x0032 099 099 000 Old_age Always - 11617995820

Device Model: Samsung SSD 840 EVO 750GB
Firmware Version: EXT0DB6Q
User Capacity: 750,156,374,016 bytes [750 GB]

9 Power_On_Hours 0x0032 099 099 000 Old_age Always - 4510
177 Wear_Leveling_Count 0x0013 099 099 000 Pre-fail Always - 4
241 Total_LBAs_Written 0x0032 099 099 000 Old_age Always - 8497497858

To be fair you've written ~950 GB and used 8%. If we simply extrapolate that you'll get ~12 TB or about 100 full drive writes.
Personally I'm also dissapointed in the consistency of the TLC NAND drives.
I get performance degradations on all my Samsung drives even though I've applied all the performance fixes.
I can't fathom though why you are putting the smallest (slowest) and cheapest drives in your server.
A Professional SSD (Sandisk or Intel) would be a lot more appropriate.

Mmm good thinking but you're assuming the swap partition is on the same drive as the system.
While this is assumption is right most of the time, so is it that the system partitions (/, /usr, ...), from which you'll be fetching that "new memory hungry application", also lie at the start of the drive.

I don't think that there's any evidence that the linux swapfile performs better - and in any case why would it being unfragmented be an advantage? Memory access is random, and so swapfile access is random, and so why does having it non-contiguous cause an issue?

That's exactly why you want a swap partition at the start of the drive where seek time is as low as possible.
You don't want a bad situation (swapping) to become even worse (swapping and seeking all around the drive).