Amazon's new M3 instances offer improved performance compared to their predecessors by providing double the computational capability/cores while maintaining a balanced set of CPU and memory resources. The M3 double extra-large instance (m3.2xlarge) outperforms the m2.2xlarge high memory instance in terms of fork time, which is significantly affected by Xen fork time issues that were previously a bottleneck for Redis performance. However, the M3 instances do not provide significant advantages in replication time and throughput compared to the m2.2xlarge instance, especially when considering the cost difference between the two. Therefore, Amazon's new M3 instances are recommended for use cases where high throughput and low latency requirements are not critical, or for running Redis without replication. For other use cases, the m2.2xlarge instance remains a suitable option due to its lower cost and comparable performance in fork time, replication time, and throughput.