I think this is done for historical reasons more than anything: if you look carefully, ReadWriteLock was added in JDK version 1.5, while SynchronizedMap was part of JDK since 1.2.
If we think in terms behavioral compatibility, it's invalid to suddenly change how synchronized map works internally (i.e. by ordering both reads and writes), because there would be a lot of programs written between 1.2 and 1.5 that would've relied on exactly this behavior. Another point to this is - the Map implementation is called synchronized, which actually does mean that it orders every access, just as synchronized
keyword does.
There's even more to this: read-write locks come in two flavors of sync: fair and non-fair. If we will replace simple mutex with read-write lock, which flavor we go for? What about different lock implementations? Etc.
It honestly makes more sense to provide a general-use concurrent map as they did with ConcurrentHashMap, and everything else will be implemented by library writers if developers need it.