Since bitexploder asked, I'll add what I wrote on this on other forums. If it's about secrets or anonymity, make sure you always use a safe language that supports careful control and reasoning about both memory and CPU time. The reason is that this enables covert, channel analysis for vulnerabilities that leak secrets through storage and timing. It's why I wanted Freenet to ditch Java aside from the obvious reasons.…
It's unlikely that the cost of whole-program side channel resistance is worth it for an application like Tor. Side-channel-resistant cryptography is almost certainly sufficient for most reasonable threat/cost models.
You can limit it to the protocol layer. High-assurance security in military systems just used fixed-speed, fixed-timing transmission with covert channel analysis of setup, crypto operations, and error handling. Any problems at all violating key invariants cause fail-safe. Tor could do something similar.
Still need a language that allows such analysis at least for those components being analyzed.