the main problem with porting nuvola tools to windows is that it uses, in part, some libraries designed for research purposes in the field of "computer vision". Since they are used mostly for scientific purposes and not in consumer products there isn't much request to port them to windows..
while it is true that many core computer vision functions needing optimization for speed and flexibility have been rewritten from scratch in nuvola tools, one especially hasn't. That's the "labeling" function, if I recall correctly it's written in C and computationally speaking, is very heavy. Therefore it can't be readily converted to a python program without an increase in processing times.
Another reason is that most data structures that allow easy conversions from and to gimp image layers, .tiff files and internal working structures are defined in specific python libraries that haven't been ported to windows as well.
Porting nuvola tools to windows basically requires a complete rewrite of many functions, extreme optimizations and probably a large RAM overhead. Not that it's impossible, it just takes a lot of time and time is the major constraint here