Let me preface my comment with a statement that this is very interesting work and it's appreciated. Nevertheless, my worry with the advice given here is that it provides comfort to those unwilling to evaluate other options: platforms or frameworks that may provide the necessary performance characteristics without as many rounds of optimization. Avoidance of premature optimization and avoidance of selection of a well-fitting platform and architecture are not the same thing, though they are often confused for one another.

If you are familiar with Django, and better yet familiar with Django REST, and do not have the time to invest in considering other options (learning about, experimenting with, and taking the necessary time to properly digest), this article gives useful data and demonstrates how to reach performance levels that may be acceptable for your use-case, with some concessions (notably, caching).

However, if you are willing to evaluate other options, there are many platforms that achieve the synthetic target established by the "third bar"--that is, 25ms of server-side processing while doing so with client-side concurrency and without resorting to caching or other potentially complicating/limiting concessions.

If you are a Django developer, by all means leverage advice such as this. If you are not, consider looking to alternatives that meet your performance target out of the box. Perhaps you will find something new that you'll eventually be glad you looked at.