My recent experiences with Solr+Magento might have a little insight.
First, there's the problems:
- Fragmented PHP libraries - Which one should I be using? How many lines of code am I going to have to fix in an unmaintained library?
- Additional stuff to learn - I had zero experience with Tomcat before jumping into this boat. It's be real fun.
- Very little documentation geared for an experienced developer in an unfamiliar environment. Be prepared to learn all that fun Java terminology just so you can read a setup guide.
- There's a lot of effort involved for the DIYer
In my case, I want to go with Solr because it does what I want. Sure some of those 3rd party search providers will do that too, but if I wanted to use that I would be the guy that is content with a $200 website. I most certainly wouldn't be trying to push the boundaries of Magento CE.
The out-of-the-box search just simply isn't going to be a good enough solution for me. I have customers that are going to rely on store search and the chances of them mispelling the names are very high. That alone is enough for me to want to get away from Magento's methods.
I think, as others have also made clear, that going down the Solr road is no light undertaking. Aside from the nuts and bolts of Solr itself, you're also going to have to worry about the infrastructure implications (something that is leading me to consider moving everything to co-location).
So far, though, it's all be a pretty rewarding experience. I imagine that it will be another couple years before I have answers to so many of the questions I have, and by then we'll be looking at the next fad that's probably running in node.js or something and it will be time to migrate all over again.