We have a customized sphinxsearch engine powering bug.gd. We can't say enough good things about it. Actually, out of all the open source stack we use, I think I'm most pleased with how robust and solid that piece is specifically. Sphinx is one of those powerful little open source gems that people don't discuss enough. It feels very feature rich and is wicked fast. I don't understand why it gets so little attention overall.
Its reindexing speed is especially good-- I highly recommend it for any service that needs their search index updated quickly. A reindex of 100,000 documents takes about a second-- just fast enough that we could actually do a full reindex on every search if we were that silly.
Supposedly you'll also find it behind craigslist's new search backend and (supposedly) the new whitehouse.gov.
The selfish inkling in me doesn't like talking about Sphinx because it feels like one of those "best kept secrets" that you'd just want to keep to enjoy for yourself-- of course that's silly.
It scales relatively well, though I wish we were getting sub-minute indexing, I'm pretty pleased with the 45 minute timespan we're seeing on a pretty large dataset.
Responses are quick, though I'm having a some problems with memory usage on my MBP, I've mostly fixed that with careful configuration.
Biggest problem is that django-sphinx is a piece of trash. I had to hack together a custom wrapper that we might consider open sourcing at some point.
It's also incredibly fast.