The risk is It will sound like the test where you bypass the dither plugin in the previous example, although the quantization noise will be much lower level and much less apparent when it’s not being exaggerated by that tool. But it will still be there at a much lower level, which you can test by rendering to a 16 bit file without dither, and cranking up the very end of a fade-out in headphones. You might hear nothing terribly objectionable, or you might easily hear the quantization noise. But It’s really not safe to not use dither, especially when rendering to 16 bit.
Back to which dither to use, it’s my understanding that the safe dither to use when audio will be further processed (which is always the case when you’re talking about a wav file you’re making that will be converted to MP3, etc. for the stores and streaming) is TPDF with an amplitude of two quantization steps.
(under “which types to use”)
Izotope has said that MBiT+, level high, no noise shaping is flat TPDF,
I’m not entirely sure about UV22.
I’m sure the Wavelab Internal Dither can be set to flat TPDF of that amplitude without noise shaping.
A user made a list of all the modes and specs of the included Wavelab dither options. If all of the information is correct, it would still apply to Wavelab 9.5.
In the real world they may all be fine. I’m sure many commercial MP3s and AACs have been made from wav files that used dither with noise shaping, even though it’s not recommended.