message queues, reliability versus speed

Hey all, been wrestling with a choice for a new project involving a few microservices that need to exchange events. We're looking at message queues like Kafka or RabbitMQ. The core issue seems to be balancing guaranteed message delivery (ensuring every single message gets processed, even if a consumer fails temporarily) with high throughput (processing as many messages as possible, as fast as possible). What are your experiences with this trade-off? Have you found specific configurations or architectural patterns that help achieve a good balance, or do you typically prioritize one over the other based on the use case? Curious to hear your thoughts!