Close
"Bad allocation" usually just means you're out of memory. It needs quite a lot of memory, as it's going to hold several copies of the image in memory at once. It probably shouldn't crash PS, but it's not well tested.

16-bit will require 4x as much memory, as it's processing it in 32-bit floating-point. This is to make the math the same as 8-bit, so the results in 16-bit are the same as 8-bit, just higher precision. Handling it as 16-bit directly makes the math come out differently and the results are different, and I don't understand the algorithm enough to understand why, or whether it can be fixed...

Color space handling is patchy; it should work, but it'll often not preview correctly or at all. Filtering a single channel in any color space generally should work, though, since it'll just display it as monochrome.