Stupid or not, the use of NT on the ship was widely reported e.g.
http://web.archive.org/web/20050214070147/http://www.gcn.com/17_17/news/33727-1.html
On a similar topic, it turns out that the G1000 uses javascript (.js objects) to some extent.
The G1000 software stack (which is Garmin proprietary, not based on a COTS operating system) includes a browser rendering component to allow OEMs (i.e. aircraft manufacturers) to customize the look and feel in some places and to provide airframe specific functionality. For example in a Cirrus, you are greeted with the image of a Cirrus aircraft. Or on a TBM there is a user interface to export trend data which is developed in HTML/JavaScript by Socata. That’s an adequate technical solution in my view.
I did actually see a piece of avionics die and show the Windows NT BSOD screen in someone’s Glastar several years ago, I was aghast that NT was being used in avionics. It wasn’t a PFD, it was a GPS (perhaps Avidyne?).
I also had reason to take a handheld Garmin GPSMAP apart (to replace the internal battery) a few years ago, it had an 80386 in it (embedded version). The Garmin handheld was from the late 90s, made not long before ARM crushed everyone else in the 32 bit embedded space.
I reckon NT (Embedded) is fine as an OS. You strip out all the rubbish like networking, printer sharing, etc… There is no free lunch in software; only time and heavy use and vendor committment to get bugs out, will get the bugs out.
A lot of Garmin stuff uses a 386. The GNS430 does I believe.
I know of a hospital where the monitors in the operating theatre ran on NT. One day an anaesthetist had a complication which should have been a non-event had he recognised it within a minute or two. Unfortunately, within moments there was a momentary power-cut – the lights flickered; the backup generator came on and the monitor decided to reboot. Had this happened a few minutes earlier or later the anaesthetist would have been able to correct the issue, however by the time the monitor got its act together the patient had died.*
On paper the anaesthetist should have been able to cope without the monitor – and indeed had he had no automatic monitoring at all I suspect he would have recognised and corrected the problem sooner. Had he had a simpler monitor based on an embedded system rather than a covert PC, the system would have probably have restarted much faster. However, one gets used to working with the tools one has to hand. Consider a pilot perhaps struggling to shift his attention to the backup attitude indicator whilst his glass cockpit is rebooting: the parallel seems a little close for comfort.
* A check with Google Scholar suggests this is not an unique event, and can therefore can be discussed as it is not identifiable.
It’s a great example but to me that sounds more like negligent hospital management, in not having a UPS on the equipment. No way will a backup generator start up in time to prevent any and every computer rebooting (potentially after corrupting its data). I am not sure of the right name for it but you can get continuous-power systems where a UPS props up the output while the backup generator starts up. For e.g. airport lights you don’t need that but you do need it for computer equipment. You can get systems big enough for a whole hospital. Without a UPS, if the backup power had taken a few mins to come online, the embedded-system-based monitor would still have been dead for that time. Still, hospitals have been around for long enough for the powers to be to have decided it doesn’t matter 
I have just timed a 2.5GHz i7 SSD-equipped laptop booting into winXP and the BIOS took 5 secs and then winXP took 25 secs to present the login prompt. That is starting up all the crap like networking, printers, enumerating the hardware and building the hardware tree for the registry, etc. I reckon a stripped-down NT would boot up in 5-10 secs on modern hardware.
Avionics can be much worse. For example AHRS can take a few minutes to restart, during which time you might have some sort of a heading but no autopilot, and maybe just the vacuum pump driven AI as the attitude reference.
One thing I find weird about avionics is that most of it doesn’t have a watchdog timer, so if it crashes, you can get a black screen, or the existing screen freezes which is even “better” because you may not notice for ages.
I also expected the hospital to have an UPS, but the lights went dark for about 2 seconds before the backup generators came on line. Doing the maths on the back of an envelope, I suspect that the cost of an UPS might not actually be worthwhile (multiply the cost of an UPS by a few hundred hospitals for the UK, then divide by the number of injuries they might prevent, set against a backup system that takes a few seconds to start up). If it cost, for the sake of argument 10 million pounds a year and only saved one life per year, then you could argue that it was harmful to install UPSes because that money, spent on road safety or staff training, might save more lives.
Every monitoring system I’ve since come across since incorporates a backup battery, but to my mind the root problem remains that the system was much more complicated than it needed to be. The measurement that would have alerted the anaesthetist could have been provided by a stand-alone device that could have taken only a few seconds to initialise had it been implemented using a microcontroller with – as you suggest – a watchdog timer. Additional more complex functionality such as the ability to print out a graph of blood pressure over the course of a procedure, could have been carried out by a computer system that took inputs from the real-time monitoring system, but on which the real-time monitoring didn’t depend.
Any updates on which avionics use which OS these days?
I use a UPS at work for the phone system and internet router, which both take ages to restart after a power cut. Despite a guaranteed minimum 5 year battery life, I make warranty claims on roughly a 1-2 year cycle. Does a G5 last the advertised 4 hours on its backup battery..?
