At my game yesterday, one of my players complained about her terrible dice rolls. She seemed convinced that they were magically biased against her. She's a statistician so I suggested she determine empirically whether her dice were likely to be biased by rolling them a lot and then use that to determine the likelihood of their bias. It's a classic probability problem. Needless to say she didn't, but it got me thinking. Given how superstitious gamers are about their dice such a test might actually be interesting to do.
The only people I've seen who seem devoted to dice innovation is a company called Game Science. I haven't used their dice though. I think I might buy a set, test them, and compare them to my existing dice.