Client vs Pool for async web servers: is multiplexing sufficient?

[richarddalves]

I'm building an async web server with Axum and trying to decide between using a single Client (cloned across handlers) or a Pool for my Valkey connection.

My understanding is that Client uses internal multiplexing, meaning a single TCP connection can handle many concurrent commands without blocking - similar to how redis-rs describes its MultiplexedConnection. If that's correct, a Pool would only provide meaningful benefits in extreme high-throughput scenarios where a single TCP connection becomes a bottleneck.

A few questions:

• Does Client use internal multiplexing, or does it serialize commands?
• For a typical async web server with moderate traffic, is a single cloned Client the recommended approach, or should Pool be the default?
• If Pool is recommended, is there a suggested starting size, or is that purely workload-dependent?