I had had to do something similar in the past but couldn’t get the distribution right. What I ended up doing was tracing from a random point on the sphere surface towards a random point in the volume.
And then finding Closest Point on Line 
It was too long ago. This created no hot-spots afaik.
There might be a better way.