I've run TweetStats on EC2 for the past year using the great scalr.net to provision new instances.<p>It's expensive as heck (the smallest instance scalr allows is an EC2 large), but it works great.<p>Scalr monitors one of several stats you can specify (RAM, CPU, queue size, HTTP response) and will spin up instances (you can specify a max) as necessary.<p>Granted the app has to be built to support such multiple app servers, which may be easier in some cases than others.<p>Sounds like the cloud server was rather inconvenient, couldn't handle immediate load, and required your manual intervention. Always tough to predict traffic spikes from Twitter. ;)
"It looked like a few hours of data was lost from the database for some reason"....eh, what? Was this the application's database? Losing a couple hours of data because of a traffic spike is not something I would find remotely acceptable.