Threaded does not imply one thread per connection. As an example, Java Servlet 3.0 allows a request handler to suspend and resume execution of a request handler on a thread. This feature allows the application to relinquish the thread for the occasional long operation and use blocking threaded code for most other code. Threaded applications written in this style can support a large number of concurrent connections with modest resource use and no callback spaghetti.

Models like Java Servlet 3.0 allow threaded servers to handle a large number of concurrent connections. Are there other reasons why threaded servlets are slow or not scalable?

Most web applications don't have a large number of concurrent connections because they application request handlers do not block for a long time. The resource use for thread stacks is very modest for these applications even without a suspend & resume feature.