If using an optimized storage driver (anything but
dir effectively), then only the first container created from an image results in an unpack. We also prevent multiple unpacks in parallel in that case.
Otherwise, you can try to throttle a bit, but it may be interesting to know where your load comes from.
The load average on Linux is simply the number of processes running or waiting to be running.
An ideally used system would have a load average identical to the number of CPU cores.
It’s also worth noting that I/O wait causes the load to increase when other processes may not be impacted as they’re not accessing the same storage device.
All that to say, it tends to be better to focus on the actual responsiveness of the system and monitor things like latency and throughput of the services running on the system, rather than looking at a pretty artificial value such as the load average.