I would master the audio the same way I would master for television. This means using compressors and a multiband limiter to reduce dynamics. If you listen to hard (dynamically) compressed pop music in an mp3 file comperessed to 128 kbit, you will not hear as many artifacts as you would if it was a classical piece. This is partly due to the reduction of dynamics, and partly due to less complex material.
For tv you can not go higher than -10dB FS. For the internet I would suggest you go a bit higher, -3 dB FS would be a good bet. That way it won't be too low compared to other internet audio, and it (probably) won't distort on bad pc/laptop speakers. But if you have to make a delivery which is going to be used for both tv and internet, I would stick with -10 dB and let the people who convert the material for Youtube normalize it to -3 dB (or whatever they use to do)
I would also listen to the mix on a laptop with crappy speakers, to see if the dialogue is clear and good compared to other internet audio.