To make it easier to see the problem:

Imagine that TrueNet has 1 person out of every 1000 subscribers.
There are two ISPs (A and B). A has 1000 subscribers, and B has 10,000 subscribers.
Both ISPs have enough cache to cache every single request for an hour, and therefore have the same
cache hit rate per subscriber.
The test is executed at a random time during an hour, and the same URLs are hit each time.
There will be at least an hour between execution times.

ISP A - 1000, subscribers, 1 tester
First run, all data is cold, retrieved from remote server.
Second run, over an hour later.
All data is cold, retrieved from remote server

ISP B - 10,000 subscribers, 10 testers
First run, all data is cold, retrieved from remote server.
Second tester, sees hot data, keeps the data warm.
...

In ISP B, at no point does the data sit idle in the cache for long enough to become cold. Does that accurately reflect customer experience? No. In fact, ISP B would have poorer results because the test is able to monopolise a piece of the cache.