I bought 6 x 16GB (DDR-3-2933 PC4-23400 RDIMM) Nemix for my 16-core, $489.99 plus tax, free shipping took two days from Florida to California despite a longer estimate at checkout. (Someone on the board was looking for 4 x 8GB, I gave him a good price and he saved me from dealing with eBay or other—we were both happy).
The most important thing is to fill six channels (either 6 or 12 DIMMs, basically the same thing). Although I've seen a lot of people add 2 x 32GB to the base (the cheapest route to 96GB), I believe that's a particularly bad way to go, leaving a big chunk of memory two channels wide. The math is pretty straight on that issue—4x is twice as efficient as 2x, and 6x is three times. Looking at it the other way, 4x is two-thirds efficient (not ideal, but not horrible), but 2x is only one-third (ugh).
At Nemix prices, if you sell the original 4 x 8GB to someone wanting to flesh out via that route, the difference in cost for 6 x 16GB versus adding 2 x 32GB to get to 96GB is minimal, and it's a lot better on memory throughput. (Cost difference is $211 plus your local tax on that amount, minus what you can get for the 4x8GB.)