I ran an analysis on a subset of the Twitter users that I follow via the streaming API. Using the REST API, I pulled the timelines of these users and compared them to what I received via the streaming API. Here are my results:
First dropped tweet was encountered on 04/13/13 ~11:00 AM Pacific
Frequency of dropped tweets increased starting on 4/17/13
Past 24 hours: 5.3% of tweets were not received via streaming API
Since 4/13/13: 3.78% of tweets were not received via streaming API
No limit messages were received while streaming, so the dropped tweets do not appear to be intentional.