When operating in 16 bit mode it definitely converts to 16 bit or the converters themselves operate at 16 bit, that’s the entire point of 16 bit mode existing - to save RAM when you use track recorders. What I’m speculating is that the converters always run at 24 bit and it’s just plain truncation without any dithering when you’re in 16 bit mode, because the coloration sounds like undithered truncation to me, but I’m just speculating about that part.
Anyway, in either mode the OT sounds OK to me but there’s definitely a very audible difference between 24 and 16 bit modes and that could be a big part of why some people have much different experience than others.