At least the Tor question should be either on the Tor site, or here on the main site. Indeed, it is a question about traffic analysis and correlation of data, which seems more on-topic on security.SE than on crypto.SE. Of course we can argue that cs.SE or crypto.SE seem related, but that's the lot of most questions on security.SE: when there is security and computers are involved, Computer Science and cryptography are never too far away.
The other one (the 2046-bit question) would be better on StackOverflow because it only calls for a generic answer which explains that the "size" of a processor is a pure question of performance (existing computers have no trouble using 2046-bit value right now; having registers with that size may just speed up operations a bit, but not significantly). Although I could engage in a discussion comparing password cracking performance on a GPU and on a CPU with SIMD instructions (a 2048-bit register would allow for 64 32-bit operations in parallel)(a good GPU would still win, though).
To sum up, I see nothing wrong in theoretical questions per se. We answer theoretical questions all the time. Answers tend to have a "practical viewpoint" (for instance, when talking about 256-bit encryption, a security.SE answer will often explain why that's overkill in practice) but theory of security is still on-topic. Which does not prevent any particular question from being off-topic, not because it is about theory, but because it is about theory of something which is not information security.