Relational database with Kafka-like durability/caching implementation

When reading about Kafka I found it really interesting how they are providing durability, especially in comparison to databases: * Kafka - never writes synchronously to disk, provides durability by replicating to other servers (and getting their confirmation); the added benefit is that it can reuses filesystem cache instead of rolling out your own * relational database - flush log to disk to guarantee durability. From what I was reading the Kafka approach results in a better performance so I was wondering if there is anything in particular that makes this approach infeasible for relational database? Or maybe some of the "NewSQL" guys are doing it and I'm just unaware? Sources about Kafka mechanisms I'm talking about: https://kafka.apache.org/documentation/#persistence https://kafka.apache.org/documentation/#replication