Placing the magnets so that they repel when the layers are aligned is really bad for a really simple reason.
When you have the polarity configuration so that the magnets attract with aligned layers, having aligned layers is a local energy minimum. There are essentially 8 local minima as you go through a full revolution: 4 when the layer is fully aligned and 4 more when it's about 45° off from aligned (exact angle depends on magnet placement), although the latter minima represent a much smaller energy gradient compared to the former and, for our purposes, may be ignored.
But if the polarities are such that the magnets repel with aligned layers, you've flipped the signs of all the potential energy formulae and now your local energy minima turn into local energy maxima. This means that the pieces will actively try to move away from being aligned. It's like balancing a pen on its tip: that's the highest-energy position, so it's not stable and it'll fall off pretty quickly without external intervention.
Cubes turn worse when the layers aren't fully aligned, so if you set it up so that the layers are essentially never aligned, of course you get a cube that turns worse. It's not rocket science.
That's not to say that making use of magnetic repulsion to reduce friction is impossible or anything; it's just that this naïve approach of flipping half of the magnets doesn't work. If you could somehow stuff some magnets and superconductors in a cube, they could conceivably repel no matter how the layers are aligned, resulting in basically zero friction apart from whatever mysterious mechanism is keeping the cubies attached to the core.
(I haven't touched electromagnetism in like six years, so my knowledge is a bit rusty and the speculative bit about superconductors is probably incorrect in some way.)
