I've built quite a few Numitron-based clocks, as well as other displays using the technology. Personally, I love them: they're easy to drive, last a *long* time, and are, to my eyes anyway, very attractive. They provide a wonderful glow, and because they're broad-spectrum, they can be filtered to any color. However I usually leave them "naked", and the non-monochromatic light they emit seems to give a richer, warmer look than LEDs, or even Nixies, to some extent. (Of course, they don't have the fully-formed character set of the Nixies.)
I've used a variety of schemes to power the devices. At first, I used a dedicated microprocessor (an AT89C2051 - about $1 each) for each tube, with the segments direct-driven from the processor outputs. I implemented a serial communication scheme that allowed the displays to be daisy-chained to any length, and controlled with a single 2400-baud serial signal. These were driven using the RCA-recommended strategy of keeping the filaments just below a visible glow whenever they were "off" for maximum lifespan, I did this using PWM, and implemented brightness controls also, since I already had the code to do the PWM. These worked very well, and the only failure I've had, even in extremely rough service (a mutant vehicle roaming the desert at Burning Man) was due to mechanical damage to the tube. (Amazingly, it even kept working for a few minutes without any glass envelope at all!)
A scheme that I've never used, but which is easy and should work well, is to use a single fixed resistor from each segment to the supply (or ground, of course, depending upon your circuit topology) to provide the keep-warm current for the filaments
For most of my clocks, I've just ignored the keep-warm function, and have seen no ill effects as a result. I've been using 74HCT595 serial-in parallel-out latching shift registers to direct-drive the tubes; their outputs are rated to sink 35 mA (74LS595 are even higher, at 130mA), and I PWM the enable input to provide dimming. I've had clocks running for a decade now with no tube or other part failures (except for wall-wart power supplies - they're often notoriously poor in quality). The highest stress is probably on the colon lamps, which are two 3.5-volt grain-of-wheat bulbs that I'm running in series to match the luminosity of the Numitrons; even though they've been blinking on and off every second for years, I haven't had any failures there either. Note that since my clocks are often placed in bedrooms, they're usually run at less than full brightness - PWM ratio of around 50% or so; however, even the one I gave to my 94-year-old mother, which runs at full power, has also had no failures.
A note about using the Numitron clocks (or Nixies, for that matter) in bedrooms: they're excellent for this application! A reasonable amount of research shows that ambient light containing a significant blue-spectrum component can disturb sleep, reducing the amount of REM sleep and overall sleep time. This is true even when the eyes are closed! The warm glow of Numitrons, especially when dimmed, contains almost no blue light, and so is the ideal bedroom clock display. I'm lucky enough to have acquired most of the remaining supply of DTF104B tubes (1" digit height, end-view), and clocks made with these can be seen easily across a room, even if the viewer normally wears glasses.
I always use direct-drive in my designs, because multiplexing pretty much obviates the big advantage of Numitrons, which is the ability to drive them directly from the +5-volt logic supply - there isn't sufficient brightness without a higher-voltage supply. I did use multiplexing once, to retrofit an existing 3-digit LED display on a musical keyboard I rebuilt into a steampunk enclosure. In this case, I simply replaced the LEDs outright with the Numitrons, and removed the current-limiting resistors. This actually worked better than I thought it would; although the displays were a bit on the dim side, they were still completely legible in normal room lighting.
Other comments about the lifespan of Numitrons are very accurate - at reduced voltages, the tubes will last basically forever. RCA data sheets show a mean life expectancy at full voltage of 100,000 hours (longer than LEDs are specified for!), and their lifespan curves are basically asymptotic to infinity at 70% voltage. Also of note is the specified ruggedness of these devices: rated impact acceleration (while operating!) of 200G, vibration 5-200 Hz 20G, etc. The Russian tubes are not specified as highly (15,000 hrs operational life - I have no idea about vibration), but I use them all the time, and again have never had a failure.
~~
Mark Moulding