For more info here, cgi.parse has code like this:

def parse(fp, ...):
    if fp is None:
        fp = sys.stdin

    encoding = getattr(fp, 'encoding', 'latin-1')

    # later on...

    return urllib.parse.parse_qs(a_str, encoding=encoding, ...)

As an easy hack, I added this after assigning encoding:

    if len(' '.encode(encoding, errors='replace')) > 1:
        encoding = 'latin-1'

I have no idea if this is a good idea or not. The current behaviour of mojibake in the parsed result is certainly worse, since the choice of utf-16-le is entirely contained within the parse() function.