My understanding (I'm a hack) is that the lack of twists causes errors in the ones and zeroes. There are always errors but the actual twist of the wires is done to minimize these. The more errors the "slower " the data. Many here are more knowledgeable than me on this, so please chime in if I am incorrect, or inaccurate.
From two different threads:
The purpose of twisting is to assure that both conductors will pick up any nearby interference (common mode), so the receiving end (differential amplifier - common-mode noise rejection) can ignore it, and pass only the difference signals. The greater the number of twists per inch, the higher the frequency (shorter wavelength) both conductors will pick up equally.
Twisted-pair works by assuring that both wires pick up the interfering signal (called common-mode noise), so it can be ignored by the differential amplifier at the receiving end (called common-mode noise-rejection).
The difference signal is passed through, while any signal common to both wires is rejected, or filtered out. The greater the number of twists per inch, the higher the frequency of interference you can assure will be ignored.