By the way, applies the same to the already downloaded pdf docs ?
yes.
It applies to everything you download and feed to an application which has internet access and which might connect to the internet based on information within the file or the filename for that matter.
For a more complete security analysis I think about it like this:
- If I download a document not over https correctly certified: the server, the last tor node and any routers between that last tor node and the server can inject something in the document - If I download a document from a server with correct https: the server (potentially hacked) could try to identify me, on top of any reservations you might have about https
By all means, that's a lot of leaks if you are concerned about your security, so it is strongly adviced to open documents in Tails or in a VM that has no internet access. On top of that, it could be difficult to verify documents and clean them if you want to store them for later use and distribution, so in that case use a clean tor connection not related to other sensitive internet traffic.
If you use tor for your everyday browsing as an extra privacy measure, than downloading a random scientific paper and opening it will probably be low risk. Just keep in mind that the last tor node is an extra MITM that makes tor under quite a few circumstances less secure than direct internet connection (since anyone can run one). So if your evince has a buffer overflow bug for example, that's an extra person who could try to exploit it (again unless you use valid https) and this sort of exploit works on any document, regardless of whether the contents are sensitive or not.
It's up to you to figure out your security needs.
Naja Melan