That's interesting. I'll have to keep an eye on Safari's memory usage, but it certainly makes sense. If it's true, it suggests that the memory leak is in the renderer, and that if the renderer crashes or restarts, it reclaims the leaked memory at the cost of all current session information.
