If that's true, then my code should do very well on it, because it uses an RLE representation of the landscape. Big areas of water are ideal for it, and it runs very fast on the big test cases in the IDE. The slowest one is "100 tests on a 1000x1000 map" at 177ms, due simply to the large number of points tested; "large map, large lake" completes in only 6ms.
Yet "No Land" in the validation suite fails for my Lua solution, and I have no idea why.
I just wrote another solution in C++, and that passes all the tests. (It uses a different algorithm based on sets instead of flood-filling.) I'm left wondering if there's something funny about that particular test case that mis-parses in Lua.