I also checked observed WS bonus per normal attack (taking the observed WS damage bonus and dividing by the number of normal attacks landed) versus weapon delay for both single-wield and dual-wield cases for linearity. WS bonus per normal attack appears to be linear with respect to total delay after accounting for dual wield delay reduction (acknowledging truncation of damage values). I'm just gonna throw the following data in here just to show I didn't make this stuff up:
BTW in the process of accounting for Fencer the TP bonus for tier V Fencer seems to be +50 instead of +55.
From regression I get -0.00001437022 + 0.00002596923*(delay). The intercept term is close enough to 0 that I would drop it, so each landed normal attack will add approximately an additional delay/385 to the WS bonus multiplier up to a maximum of 1.3. This assumes the bonus is independent of the number of normal attacks contributing to the total WS bonus. My understanding is that Ravager's Mufflers +2 doubles the bonus but the 1.3 cap cannot be exceeded (these findings suggest this).
Remember that if you have the average number of attack rounds per WS, you need to multiply this by 0.95 to get the average number of normal attacks that actually landed (missed attacks do not contribute to the Restraint bonus) to estimate the average Restraint WS bonus.