This does not only apply to gerrymandering but any way to skew a voting system.
To see this, let’s first make a few simplifying assumptions:
- We have a pure two-party system.
- Votes somehow assign seats to votes in some parliament.
- The status quo is an equally divided parliament (see this answer of mine, why this is a generally valid assumption even in an unfair voting system).
- The goal is to have the majority in the parliament.
Now let’s look at the general number of seats a party receives in dependence on its votes.
Mind that these are generic curves, not reflecting an individual election.
If you so wish, this smoothens the noise caused by effects of individual elections, e.g., candidate-specific specialities, short-lived hot topics, etc.
In general, this is about the typical case – you can always find a particular (atypical) voting outcome where effects are reversed.
In a proportional representation system or even an ideal plurality voting system, the portion of seats received by a party is identical to the portion of votes (1).
Now, suppose our party changes the system to their advantage (2), be it via gerrymandering or some other manipulation.
In our scenario, changes of power happen when we cross the 50% seats (blue line).
In the unfair system (2), this crossing happens with a steeper slope.
This slope quantifies how prone our system is to landslides:
The steeper the slope, the bigger the effect of a single vote.
Thus, any skewing of the voting system makes the system more prone to landslides.
Now, you may ask what if there is a dent in the curve causing the 50% line being crossed at a smaller slope (3).
However, such a curve is usually not generic in the above sense:
A slight change in the political landscape, regional effects, etc. may easily shift the dent in the curve up- or downwards (thus increasing the slope).
For this curve to be generic, there must be some weird mechanism in the voting system that ensures that the dent is always at 50% seats.
Now, let’s turn to gerrymandering.
If our party used gerrymandering in its favour, we can specify things a bit (2 → 4).
We have two strong increases of our party’s seats in dependence of the votes it receives:
One at low percentages where it wins over the cracked districts and one at very high percentages when it wins the packed ones.
However, this does not change that the 50% line is crossed at a higher slope than for a fair system (1).
For ideal bipartisan gerrymandering, we have a different situation (5):
Now, we have a decreased slope at the 50% crossing as the current seats are relatively secure for either party.
If we take this as it is, the chance of landslides is decreased.
However, it is questionable if this is really generic (like for Case 3).
If the political landscape slightly changes, the 50% crossing moves up- or downwards and thus to a region of larger slope.
This is an inherent effect:
If you want to decrease the slope at any point, you have to increase the slope elsewhere – the average slope must be 1.
Thus changes happen more abruptly – if they happen.
And from this point of view, our initial assumption that this about getting the majority is not necessary anymore.