Here are some issues that I've seen becoming important when scaling to a higher traffic with both MySQL and PostgreSQL (we're running both for different kinds of databases). Some of this is second-hand from my colleagues, so it might not be exact:

* Thread-per-connection (as opposed to process-per-connection with PG) means much lower memory overhead with many concurrent connections; with PG you might need to use something like pgbouncer, which doesn't come without its own issues.

* Logical replication; the MySQL format is compatible between minor release versions (e.g. between all 5.x releases), which enables upgrading individual replicas or master without complete shutdown. With PG this often a becomes way more complex.

* Better caching/buffer management. Essentially, InnoDB has its own implementation of LRU cache while PG relies on kernel page cache. This generally allows finer tuning on the database configuration level and you don't have dive into kernel page cache tuning (which will be always OS-specific).

EDIT: most of this stuff is probably even better described in the Uber article, linked in other comments here.