A good way to get even light is to have a constant LED density for all words. One simple way to do that is to use one LED/character. So a eleven-character word lights 11 LED. This would also make it simpler with diffuers.

Next thing - what does it matter if you can do it in 1kB? The majority if 8051 chips have 4kB or more.
But 11x10 characters seen as 11x10 pixels is 110 bits or 13.75 bytes. 12 hours*14 bytes + 12 minute states*14 bytes is 336 bytes. So it should be possible to do it in less than 1kB with a LED/character too. Adding a couple of extra lines of code, the lookup table can be shrunk even more.

Using multiplexing or not really isn't a problem. Tiny code and trivially small. The processor is hardly likely to need the power for something else, so it is irrelevant if it refreshes the display or if you have a statically driven display. And the power for driving the LEDs will make it irrelevant if a multiplexed display makes the CPU process some extra instructions. Just because you would not like to solve the problem with a LED matrix is no reason to consider it a non-applicable solution. It really is a well working solution. And guess what? You would not have to rewire any PCB and rearrange any LEDs depending on language.

From earlier discussions, you would know how easy you can dim down the intensity during the night. With each series resistor only controlling a single LED at a time, you would not need to find different resistance values depending on the LED count for a word.

Making use of one extra signal and buying dual-LED modules would allow red or green selectable. Two signals would allow red, green or yellow.