* Why don't you think the optimizer is a consideration? That's a big deal, because it helps the database choose the right algorithms as your data changes. Continuing gracefully while the input data is changing sounds like scalability to me.
* Postgres is generally considered to scale to many concurrent connections and many cores. Useful for OLTP.
* Sometimes the right feature or extensibility hook can make a huge difference allowing you to do the work in the right place at the right time. For instance, LISTEN/NOTIFY makes it easier to build a caching layer that's properly invalidated -- not trivial to get right if the feature is missing, which might mean that you're not getting as much out of caching as you could be.
Tuning for SSDs, compression, buffer caching, etc., are the antithesis of scalability. Those things are pure performance issues. Scalability is about a system that adapts as input data and processing resources change. (Your point about replication does concern scalability, of course.)