Even if you make the LEDs perfect diffuse emitters (which is hard to do at decent efficiency), you're still going to be fighting the infamous cos**4 law, which makes the illumination of a surface from a given source fall off very steeply with illumination angle (measured from the surface normal). The four powers of the cosine come from three effects:
- Even perfectly diffuse sources fall off with increasing obliquity, basically because they look smaller seen edgeways. Flat surfaces foreshorten as the cosine, concave ones faster, convex ones slower (and spheres, not at all).
- Whatever light you do receive gets spread out over a bigger area--evening shadows lengthen as the secant of the angle, so the intensity (power per unit area) goes as the cosine.
- The distance to the source goes as the cosine, and the intensity goes as 1/distance**2, so that's another two powers of the cosine.
Thus you get some pretty bad scalloping unless the light sources are spaced close together. Using cosine**4 as an approximation will let you calculate how bad it's going to be.
Note also that once the individual sources are frosted, putting other scatterers in front won't help as much as you might think. You do win some because the light spreading out sideways reduces the nonuniformity from the cos**2 falloff due to radial distance, but you'll win a lot faster by using more LEDs.
Cheers,
Phil Hobbs