quay uses number of cpus to scale workers: https://github.com/quay/quay/blob/master/util/workers.py#L22

this is wrong when running in a restricted cgroup/as a container/on the same kernel as other unrelated processes, this caused PROJQUAY-133 on our environment.

At least reducing number of workers for smaller quay containers helped us to get rid of "413 Request Entity Too Large"/"writev() failed (105: No buffer space available) while sending request to upstream"