Issue with multiplicative holiday effects

I struggle with holidays in combination with using the multiplicative mode.

I have daily data, for about 5 years. In my series, I have a 1 day event which massively increases the value of my y variable:
typical value y (unscaled): 0.01, value during event day: 0.25. This same event happens 2 times in the training data.

When I include this day as a custom holiday, and using the mode _additive_, the predictions take into account this event, with a fitted effect of about +0.22 on the event day. So far so good.

When I switch to _multiplicative_, the estimated (unscaled) effect is 0.04 (so simplifying, it is about an effect of ((1+0.04)*trend), so this is a very small effect and the fitted y value is way too low on these event days.

Is there any known issue related to using the multiplicative mode, or I am I overlooking something?

Component plots side by side:

**Relevant settings:**
Seasonality_mode='multiplicative' (documentation states that this also means that holiday effects will also be modeled as multiplicative: https://facebook.github.io/prophet/docs/multiplicative_seasonality.html)
Holiday prior scale: 10
yearly_seasonality: [10, 15] (tuned)
weekly_seasonality: [3, 6] (tuned)
seasonality_prior_scale: [0.01, 10] (tuned)