This is happening because common RGB LEDs, from a colorimetric perspective, suck.
The three elements in each LED part are positioned fairly far from each other, and they are neither equally spaced nor uniformly shaped, so you end up with strange offsets and banding through interference patterns. With a single white emitter you'll still get fringing effects (e.g. chromatic aberration) due to the varying refraction angles of the different light colours and the LED not being a point light source, but when the primary colour emitters are spaced apart the effect is even worse. This is particularly bad on your typical WS2812B in a 5050 package, because the spacing is large. The smaller 2020 package ones are better, but far less common. The problem is exacerbated by variances in the radiation characteristics (viewing angle) causing differing intensity attenuations at oblique angles. Rotating the LEDs won't work very well - you'll just create different interference patterns.
This problem isn't as bad when you're just looking at the LEDs from a distance, because diffraction, diffusion, dispersion, and defocusing effects take over and "blur" the three emitter outputs into one, but if you look carefully you can still usually see that slight off-white effect around the edges of each LED when a low saturation colour is displayed. Putting a surface very close to the LEDs breaks this illusion because the interference effects dominate at that short distance.
The quality of the "white" that you get from turning all elements to 255 is also horrible on RGB LEDs - I don't know that I've ever seen anyone calculate the CRI, but I'd bet on it being somewhere around 50-60. If you need the white to look good, e.g. for photography, you'll want to use proper white LEDs with a CRI above 80 (ideally above 90) with a known colour temperature so you can set your white balance.
One solution is to instead use RGBW LEDs, and ideally smaller LEDs. The closer the grouping of the emitters inside the package, the less of a problem you'll have in short-distance applications like this. The white LED will dominate the light output and, since it's a single emitter with a fairly broadband output, it won't exhibit the same Moire-like patterns with multiple emitters. Since the individual RGB emitters in an RGBW LED are then typically driven based on a subtraction of the minimum intensity level (i.e. the lowest RGB channel value is used as the white level, and the RGB levels have that number subtracted) the undesirable effects from emitter offsets become less pronounced.
You could also try to mask the issue by moving your translucent diffuser away from the LEDs. This causes the optical power density to even out a bit before it hits the diffuser, improving its overall effectiveness at spreading the light. Whether or not you can do this in your specific case depends on the mechanical constraints you have.
