I think I've discovered the best way to smooth over the liquid tiles. It does require a fair amount of extra tiles to be generated, but it shouldn't be any slower than the current system.
I did a bit more looking into the state of Python. There are some very promising avenues on the horizon for it, that should be of interest to other Python game developers:
1. There is an optimizing compiler for Python coming up called, in the style of Monty Python references, Unladen Swallow. It's developed by Google so it has some strong support behind it, and the word is that eventually it will make it into the stock CPython distribution. That, coupled with the existence of Cython, make it clear that there is strong hope for dropping Psyco in the future.
2. There is some murmuring in the community about creating a version of Python for iOS. That's a target platform for me. I don't know what the performance will be like (word is that it's not bad), and anyway the "gold rush" phase of iOS development seems to be over, but it's still a bit of a vanity thing for me to make an iPad version. We shall see.