For near-real-time streaming workloads that are highly latency-sensitive, the network layer must provide ultra-low latency and high throughput between compute nodes. Elastic Fabric Adapter (EFA) is specifically designed to meet these requirements. EFA enables EC2 instances to communicate using a high-performance network interface that bypasses traditional TCP/IP networking, providing lower and more consistent latency. This is especially valuable for tightly coupled workloads that require fast inter-node communication.

Option B is the correct choice because EFA supports custom networking stacks and is optimized for applications such as real-time data processing, distributed computing, and high-performance workloads. EFA integrates directly with supported EC2 instance types and allows applications to take advantage of enhanced networking capabilities without requiring complex multi-VPC architectures.

Option A introduces additional network hops and latency due to VPC peering and is not suitable for ultra-low-latency workloads. Option C (spread placement groups) is designed to increase fault tolerance by placing instances on separate hardware, but it does not optimize for low-latency communication and may actually increase network distance between instances. Option D improves storage performance, not network performance, and does not affect inter-instance latency.

Therefore, B is the best solution because Elastic Fabric Adapter is the AWS-recommended approach for achieving the lowest possible network latency between EC2 instances in performance-critical streaming architectures.