It started as Lucene, but is quite different by now as Jim (http://friendfeed.com/jim) had to create all of the real-time indexing and standing query support (at any moment, it is simultaneously executing thousands of queries).
You don't need special Google sauce for plain text search. The tricky bit of something like this is scaling it out to where it's instantly indexing every new bit of text that comes across, but that's still, while a decent bit of engineering work, is not a conceptually terribly hard problem.
It appears to be accomplishing this using tight polling (from my investigation with FireBug). Interesting and useful, but not exactly innovative from what I can tell. Also, no way to pause or stop it?
I'm not sure what you mean by "tight polling", but we basically keep a connection open to the server at all times (often known as "long polling"). This is similar to how IM works.
You can pause it by pressing the pause button on the top-right of the feed.
Ah, I see you can pause it only if you are logged in.
The jsonp requests I was seeing in FireBug did not exhibit the normal kind of long polling behavior I have seen on other sites, so I might have been wrong. Apologies.
What's interesting is that new posts are added to the search index in real time. Twitter used to have this (Track) but couldn't scale it; they're testing it and may reintroduce it. For FriendFeed to have put this out I think they must be confident it can scale.
Friendfeed does have to think about it compared to Twitter at the time they removed Track, more than a year ago. Even then Twitter had more traffic, but it wasn't nearly as dramatic a difference as in the more recent part of that graph.
