I am currently wrapping up a project whereupon I am adding pg support to a previously mysql-only large open source application. The footguns that I discovered when doing this (this is a very CRUD-like web app) weren't performance related, but rather the bizarre data integrity issues MySQL ships with out of the box. Did you know, for instance, that the default collation that MySQL installs with is case insensitive as well as accent insensitive? This means that data that contains é, for instance, will match the e character in queries. MySQL is in general much more forgiving about data type coercion than Postgres as well. MySQL makes assumptions about certain things regarding dates and times that Postgres doesn't.

All of the above bodes well for someone trying to quickly hack together an app that does stuff, but for long term data integrity it's much easier than pg to introduce subtle differences and flaws in the data through some application code that works at face value for CRUD operations out of the db but down the line fails due to subtle data differences.