[capableweb]

Load as in load average? Or load as in traffic?

Spikes in request latency can be because of bunch of stuff, including more traffic, but in my experience, it's usually around non-existing optimizations for some data structure that got triggered after N items or new deploys containing code that wasn't as optimal as the author of the code thought. Especially when dealing with distributed systems, where sub-optimal code in one part can cascade performance issues to various parts in the system.