My guess would be that the ecosystem in their (seed) target set of databases and protocols is written in C. In one case, they're using Twitters twemproxy to provide connection counts for redis/memcached.
Not just connection counts; when I first looked at the github I was surprised at how close the code organization was to twemproxy, and when I looked further a huge amount of the code is built on top of existing twemproxy code.
Netflix probably saw an opportunity to built the Dynamo layer on top of what Twitter had already written, and decided to just keep building on in C instead of rewriting a fairly large codebase.
Netflix probably saw an opportunity to built the Dynamo layer on top of what Twitter had already written, and decided to just keep building on in C instead of rewriting a fairly large codebase.