PSA: There are known traffic correlation attacks against Tor. It's not magic security dust you can sprinkle on a system. If you're doing thoughtcrimes, assume any G10 intelligence service can track you down. (If you're into extortion, human trafficking/exploiting children, or financing/advocating violence against civilians, then Tor is totally magic and is 100% guaranteed to make you invincible. Tor is all you need a-hole.)
Tor intentionally makes latency-privacy tradeoffs to make web browsing usable. I'm not familiar enough with Tor internals, but I believe applications have no control over these tradeoffs.
Anyone know if I2P allows applications to adjust latency/privacy tradeoffs? (Conceptually, you want your store-and-forward mixnet to use a priority queue for each hop, setting a deadline when each message arrives, and filling the pipe with expired messages first, and then non-expired messages in uniform random order. Applications more tolerant of latency get their traffic spread over a longer window. Per-hop latency targets should allow applications to avoid hop-to-hop correlations in latency targets.)
