Since this is too long for a comment, I'll post it as an answer.
easily legible ... known reference ...
Visually, you (better yet, your team - at least one of which is colorblind) should make the determination. That is because of many factors - LED efficiency, batch / vendor differences, LED degradation over time, human perception / acuity differences, temperature effects, ambient changes (direct sunlight versus fluorescent versus dark), the list goes on and on. I've tried meeting standards and results were lackluster at best; always ended up deviating. Also battled with repairing/adjusting devices which didn't care about standards at all (blindingly bright at night and barely bright enough to read in room lighting.) It is a good topic to carefully consider.
You can try to adhere to luminosity standards, but ultimately they are moot in the face of all of these variables. All the hard work using them goes out the window when the user places your device in an environment you didn't envision (like in an airplane with moving sunlight and shadows), or when the piece of red plastic over the display must be sourced from another vendor and the new window suddenly makes the display seem way brighter (or dimmer) than before.
I'm not sure an adjustment pot is the ideal answer here, but it's not a bad idea for the unforeseeable. And an ambient light / photo-detector to servo the brightness is a really good idea if expecting to operate in wide ambient range. The range from dark to sunlight is logarithmic and vast. Could even use an LED and detector internally to compensate for aging if desired. (Have an LED alarm clock from the 80's that still runs (and servos its own brightness), but is quite dim these days - only way it could have been made better would be to compensate for aging.)
PWM ... [Persistence of Vision]
Researching Persistence of Vision will get you going on this topic. The books in the comments are a good idea as well. They will say that about 20 updates per second will "persist" enough to seem continuous. Perhaps they were talking about full-frame motion pictures (maybe even interlaced) but this certainly is not the lower limit for character displays. An update rate of 100/s is still detectible, especially when the eyes or display moves. Many people dislike flickering of any sort. Like allergies, some are much more sensitive to it than others. The faster you can update it, the better. Might make sense to dedicate one micro just to displaying the data quickly with accurate timing, then pipe it the info to display over I2C or SPI.
Also consider matching integer camera shutter/video speeds if you care about the photograph-ability of the display. Interesting experiment - use a cellphone to record video of the (character display) instruments in a private airplane cockpit. Since they were not designed for this, characters will be totally unreadable on video.