Interesting idea. Mathematically it seems that it would on average create a perfect distribution. But it'd also lend itself to some substantial variance.

Imagine you have a 650-350 voter lead in a state with 10 districts. The way you maximize the number of competitive election is by packing 3 districts 100-0. The party with 650 voters now has the same number of voters remaining as the party with 350, for the next 7 districts.

Mathematically this works out perfectly because the party with 650 voters has an overall seat expectation of (3 + 7 * 0.5) = 6.5, but if you run the math further here's how it looks. The party with 35% of the representation gets (assuming a 50% outcome per competitive election):

===============

0 seats = 1%

1 seats = 6%

2 seats = 16%

3 seats = 27%

4 seats = 27%

5 seats = 16%

6 seats = 6%

7 seats = 1%

===============