This is likely a science project, but we could get some fairly large gains with respect to traffic analysis if we padded packets. While timing and number of packets in flights likely leak a lot of information that simple schemes won't remove, a small amount of obfuscation doesn't cost a lot and could help more precise targeting of analysis techniques.

Note that padding non-ack-eliciting packets requires careful attention to the effect this has on the congestion window.