He wouldn't speedBLD in MBLD. No one would.
But you could use more efficient packing of information. Example: if you want to memorize permutations of the numbers 1 to 4, you can of course memorize it as four digits like "1432". But there are only 4!=24 possibilities, so you could instead encode the permutation as a number from 1 to 24. That's just one or two digits to memorize instead of four. Or one letter, if you prefer.
Similarly, you can memorize a 3x3 as a string of about 20 numbers from {1,...,24}. But 24^20 is about 4*10^27, about eight digits more than the number of actually possible cube states. So you're memorizing eight digits more than strictly necessary! You can for example break the cube state down into EP, CP, EO and CO and represent each of those with a number from 1 to 12!/2, 8!, 2^11 and 3^7, which would be
22 digits. Of course those numbers are still quite large, but for 2x2 some useful tighter packing might be possible.
Btw, Richard Carr worked on a method for
solving 3x3 blindfolded in pretty much 4-5 algs: one alg each for EO, CP and CO and one or two for EP.
