... they tend to choke on the "all 0s" seed (with modification this can be a different value).

There are modified designs, but I would not bet on their uniformity.

I'll try to come up with something which has support in downloadable literature. Of course I am no cryptographer/matematitian, and am not going to perform the Diehard test suite and similar on any quickly concocted scheme.

Of course the block cipher based pRNG is as good as the block cipher is; the only problem is that it is slow. If this is important or not, depends on the application. Although XTEA did not receive that amount of cryptoanalysis than "adult" ciphers such as AES or DES, it still seems to be solid enough.

JW