I did an experiment with a very simple cache based on a simple concurrent hash map. This showed me two things:
Caching existing resource URL gives almost no improvement (maybe some, but it seemed to be more noise than anything else).
Caching also non-existing resource URLs indeed gives a big improvement, the load time seems to be down to approximately the value of 15.10.16. I didn't measure often enough to be able to truly say if it is the same or not, there was too much noise in my tests. So I think this might indeed be the root cause.
I don't know about all of those numbers as I didn't want to manually test all of this - it would really be nice to have a script that prepares the wiki, makes sure that all background tasks completed, and then performs all those measurements and generates a report.
This message was sent by Atlassian Jira (v9.3.0#930000-sha1:287aeb6)
If image attachments aren't displayed, see this article.