That's a lot different from my testing, but I'm sure it's valid. I just took another stab at the memory leak issue, assuming it has to do with screen transitions, and by removing part of the transition code, the memory leak was slowed down in a huge way. It didn't get rid of it entirely, but it looked pretty promising. I haven't yet had time to investigate further, but if this doesn't allow me to actually fix it, I might be able to add an option to truly disable screen transitions, and tackle it that way, at least for the time being.
FYI, screen changes always occur using screen transitions, even if they are disabled. Basically, if you disable screen transitions, it's still the same code path, it's just that the transitions are "instant". That makes the code easier and more consistent. I might just code in a separate path that doesn't use the transition code. Anyway, more investigation is needed...