I basically don't read anything online except through RSS feeds. My strategy for dealing with high-update sites is twofold. First off, I don't subscribe to redundant feeds (since many sites offer highly-overlapping content), and I am very trigger-happy on the unsubscribe button.

Secondly, I make use of this highly customized neural network-esque technology I use which gets a finely-grained history of everything I do and uses it to build a model of importance by which it can filter out irrelevant content. It's a bit processing intensive--it can only filter through 30-50 items per minute--but it's not a big deal since my RSS feeds only produce less than 500 items a day, so that's only 10 to 20 minutes of processing per day. It produces (on average) only a couple of false positives (items I don't actually care about) per day, and (although of course it's hard to empirically confirm) it seems to produce similarly few false negatives. This is a huge gain over every other system I've tried (especially populist voting-based systems).