One the oft underlooked benefits of it's progressive decoding implementation is you can just dynamically truncate the bytestream to serve lower res versions, no generation of alternative sizes in advance required.
For the Modular file, it failed to partially decode it. This was supposed to be a major feature of FLIF, and it's not working here at all.
For the VARDCT file, it made the top 2/3 clear (looking like a low quality but full resolution image), while the bottom 1/3 was extremely blurry (as if it only had the DC values for the blocks). When I gave it 5/8 of the file, the whole image looked full resolution with no bad blurring, but was still low quality. But roughly the same quality level than what the image would have been if it had been saved at that size.
So it does work for VARDCT files (and you must use the -p switch when saving it), you just need to give it enough bytes. No clue what's going on with the Modular file though.