It isn't mentioned here, but hashes are a big win in terms of memory savings. The hash representation for small hashes is very efficient compared to keeping top level keys. You can get savings of around 25% by moving data into hashes with five fields, instead of having 5*n keys. It's made me drastically reconsider how I've been using Ohm, because now I want to encode my models into hashes.

I assume you've read it already, but for anyone looking for further reading on Redis' hashes, see last week's update:

http://antirez.com/post/redis-weekly-update-1.html

Not just 25% of memory saved, the same data will use just 25% of the memory needed before hashes.