The biggest differences normally arise with multiple queries running concurrently. For example, InnoDB & PostgreSQL use MVCC whereas MyISAM uses locking, and PostgreSQL normally wins most benchmarks with high concurrency on multi-core machines (8+).<p>I'm not certain I understand what you ran in the 'heavy' and 'light' threading tests. Can you provide some more details / explanation?
Everyone measures how expensive transactions are compared to no isolation, but I've never seen anyone switch off isolation and measure how often their code stumbles across a half-baked update from another thread, nor how badly their code screws up their data when it happens.
My problem with PostgreSQL is that you're half expected to tweak it to your needs. Ideally I would have to record and replay real life activity against the database for each adjustment. It's too much work to do properly. An expert could probably make a good first guess though. Which is why I've wondered if there is a tool where you drag a couple of sliders that describe your server specs and application nature and output config.