With the recent work on the upgrade tests I changed the counters testing there a little bit.
Now the test creates 100 separate counter cells. These are updated randomly for 15 seconds (so we'll get sorta even distribution of counts across the 100). Then the counts are read back and compared to expected values. On some versions I was running into very slow counter updates which is why I opted for a max time approach instead of a fixed count, though in retrospect maybe those very slow scenarios are worth investigating more.
Aleksey Yeschenko does this update seem like reasonable coverage, or should we extend it a bit more? Let me know if there's any specifics you'd want to see covered in a test.