Specifically: "this is the first patch where we've been unable to keep the non-memcached version as fast as before. (Well, we could've, but it would've involved massively parallel codepaths, one of which would've bitrot.) So if you're running a loaded LiveJournal-based site and plan to upgrade to this code, be sure you're using memcached."
This code is live on LiveJournal, so you can keep an eye out for bugs, most of which get reported to the Support board over the next few days - a million users sitting on their friends pages is the number one way to regression test code.
You may wish to look at speed differences, but brad has indicated that the new code may even be faster. Unlike how I read it, it will work just fine without memcache - I can't read tonight and had a totally silly idea in my head, and I'm not sure why, so I apologize for the harsh and (most likely) incorrect warning. Just something to keep an eye out for yourselves.