That's an interesting approach for dice, and one I had not seen before. It is essentially the same basis as the Von Neumann debiasing algorithm for coins I linked to earlier in the thread, where you flip twice and discard the result if the two flips are the same, but adapted for dice.
Having said that, it's also far more complicated. I like to keep things simple and would just stick to coin flips.
Are these really biased?
You would never know unless you rolled them thousands of times and performed some statistical analysis on the results. Safer to just assume they are and use a debiasing method.
And I would roll it 256 times anyway, and treat each value as a single zero or one, to avoid modulo bias.
You still need to correct for physical bias, which is far more simple with a coin than with a die.
Flip the coin twice:
HT - 0
TH - 1
HH or TT - discard
Repeat until you have 256 bits. No matter how biased your coin is, you'll always get a completely random result.