I'm scared of your brain.
Exposure therapy - Wikipedia
This is actually funny because if you switch the direction of the U/U' from current scrambles (i.e. always use D++ U' and D−− U), then it actually improves scramble quality a bit.The U/U' is dependent on the previous move
All 2/5 turns is (much) better than all 1/5 turns at mixing up pieces on a megaminx, but it turns out that introducing a small fraction of 1/5 turns is good too. (There's a huge edge orientation bias when fully restricted to 2/5 turns; it takes only three 1/5 turns to flip an edge but five 2/5 turns to do the same.)
(Changing the U direction is a fairly small improvement and I have alternatives to Pochmann scrambles that are much bigger improvements. Still working on a writeup for this!)
This is somewhat veering off-topic, so spoiler box:However, will you always see a new scrambled position? More than likely. […] In order to see every scrambled position of the mega, you would literally need to live and stay awake for years longer than possible. I think that the WCA scramble works fantastically. I will admit that during at home solves, I usually do a few extra moves to separate stuff in order to make it look more scrambled.
Merely having a large number of possible positions isn't good enough. It does (practically) guarantee that repeat scrambles won't happen, but that's about it.
Consider ⟨U, L, F, R, BR, BL⟩. These six moves will cover around 2.0e42 positions (comparable to two WCA scrambles, with 2^140 ~ 1.4e42 scramble sequences), but they're all extremely heavily biased towards being easy – after all, they guarantee F2L skip every time!
So the real question should be: do WCA scrambles exhibit noticeable biases? For example, are free pairs too common? Unfortunately, the answer is "yes" (e.g. corner-edge pairs are ~33% too common, and as mentioned above, flipped edges are too rare), and it takes many more moves to reduce these biases to an unnoticeable level. (120 or more moves, depending on what exactly you measure and what you consider "unnoticeable".)
What you shouldn't do is to do a normal WCA scramble, then look at the puzzle and deliberately break up pairs. This introduces human biases! Instead, try either of these:
1. Increase scramble length. 100 moves still doesn't take that long to do. (Doesn't fully fix the problem (the biases are still detectable), but it's a start.)
2. Deliberately break up pairs, then do a normal WCA scramble.
Consider ⟨U, L, F, R, BR, BL⟩. These six moves will cover around 2.0e42 positions (comparable to two WCA scrambles, with 2^140 ~ 1.4e42 scramble sequences), but they're all extremely heavily biased towards being easy – after all, they guarantee F2L skip every time!
So the real question should be: do WCA scrambles exhibit noticeable biases? For example, are free pairs too common? Unfortunately, the answer is "yes" (e.g. corner-edge pairs are ~33% too common, and as mentioned above, flipped edges are too rare), and it takes many more moves to reduce these biases to an unnoticeable level. (120 or more moves, depending on what exactly you measure and what you consider "unnoticeable".)
What you shouldn't do is to do a normal WCA scramble, then look at the puzzle and deliberately break up pairs. This introduces human biases! Instead, try either of these:
1. Increase scramble length. 100 moves still doesn't take that long to do. (Doesn't fully fix the problem (the biases are still detectable), but it's a start.)
2. Deliberately break up pairs, then do a normal WCA scramble.