We do all our operation over wifi and this basically forced me to rewrite all the device drivers with roscpp on my systems (in particular axis_camera).
The bug is even more subtle than that. I can't prove it right now, but I think if you start listening to a Python publisher over wifi, then stop listening, send the robot over a long loop where it loses connectivity, and then start to listen again when the robot is back, you will observe a few seconds of delay in your data. That happens for small data (e.g. compass) or large chunk (e.g. images).
For images, without moving away, then delay can accumulate slowly: I've done face tracking demos over wifi with a PTZ, it started fine and was basically unusable by the end of the session due to accumulated delays. Restarting the listener would not solve the issue.
rospy in its current state is now used very very sparsely on my embedded systems.