Slightly tricky question to answer solidly. Microsoft stipulate that the larger 18byte WAVEFORMATEX structure should be used for all non-PCM and compressed formats (which in practice includes floats); and all their examples show this. But as with so many aspects of RIFF, documentation is scattered and piecemeal. The floats format was developed in tandem by Sonic Foundry and Microsoft, and much of my information actually comes from the former.
More to the point, most commercial applications bend over backwards to open a file if they can - I have had files with known errors in the header (wrong data sizes, etc) be accepted by both Microsoft and third-party applications. On reading, of course, you can parse for both 16byte and 18byte format chunks, and this is what I do, and the majority of applications would do. The more interesting question thus would be if Samplitude will accept a floats file with the 18byte format chunk, which it should. But it is simply churlish, and no real help to the user, to be dognmatic on reading a file, when there is actually nothing to stop you accepting it. So the fact that an application will accept a format is no actual guarantee that the format is 100% correct - unless that is actually the stated purpose of the application in the first place.
So my answer is a bit Orwellian; both 16byte and 18byte chunks may be 'acceptable', but the latter is more correct than the former!
Richard Dobson