I did some testing on this a few years back and found that if you are wanting to simulate light behind a translucent diffuse material, it was really better to make the diffuser itself emit light, rather than trying to emit light from behind it.
Take a look at these two images. The first image uses a point lightsource inside the box - with a diffuser made from a squashed cube with a white translucent shader applied.
The second image - the light was removed and the shader on the diffuser was changed to be a white emission shader instead.


There is very little to tell them apart. There is a slight falloff towards the edges of the translucent shaded diffuser, whereas the emission one is completely uniform. This falloff could be easily simulated using a radial gradient however.
The biggest difference between them is performance. To get a comparable level of noise between the two images - the one with the translucent diffuser took 4x as many samples to render (600 vs 150) - so there is a massive performance gain to be had going down the emission shader route.