The problem appeared to be a lot of unnecessary disk io coupled with DNA lookup that only get done after every single read request is complete. This means that when tab #10 is taking long to read whatever from disk it blocks every other tab.
Noticeable only when using spinning rust disks.