Here’s what I think you’re asking:
Within a 10 minute timeframe,
In this case, both Joe and Mary would have connected to gmail.com over the same Tor circuit.
Restarting Tor Browser deleted all application-level data so, for example, Joe and Mary would have had different cookies.
You can watch this process yourself by installing onioncircuits. It’s a great tool to learn about stream isolation.
This is a very concerning issue. I can imagine someone creating a bitcoin wallet at blockchain.info. Then, transferring bitcoins from wallet1 to wallet2 via some anonymizing service. Then, closing Tor Browser and logging into wallet2. While the bitcoin trail might have been obfuscated, both wallets would have been accessed from the same source.
[This example is complicated because blockchain.info doesn’t allow clearnet connections from Tor exit nodes. But I confirmed that the connection to the hidden service occurs over the same Tor circuit as well. However, I don’t know enough about hidden services to know what identifying metadata can be collected regarding the source of the connection - meaning IP addresses aren’t used, but can the relay be identified as being the same by it’s fingerprint?]