Is event-driven architecture slower than regular request-driven architecture?

I just learned about event driven architecture and it's quite neat, though I have a small gripe with it. Every time you publish an event, it gets written to disk to be persistent. Compared to requests, this disk IO seems like it would introduce a lot of latecy and bottleneck. In actual systems, are event-driven architectures observed to be slower due to this?

On another note, is there such thing as an in-memory message queue (one that doesn't write to disk)? For niche applications where it doesn't matter if 20% of messages don't make it but you want it to be incredibly fast to handle frequent events or to minimize load on the rest of the system, for example.