Usually, we introduce subtasks if we want to avoid a probable gap between problems. For example, here C1 made the round more interesting for ~3000 participants. Strong issues would not have appeared if it was absent, but don't forget about thousands of participants for whom D+ were too difficult. Similarly, the gap between F and G2 is looked to be huge, and I've introduced G1.

In a perfect world, probably, subtasks are not needed. But in the real world, sometimes, it is not easy to estimate problem difficulties. Also, problems are not snacks — you can't easily get new or replace them with some extra. I see that a subtask is a method to make difficulties smoother, to make round interesting for more participants. Without subtasks imbalanced rounds will appear more often.

This time I overestimated G1, it should cost less. But I think that easy versions should cost reasonable scores. Say, here 2000+1500 would be much better. For example, I see that C1 and C2 cost reasonable.

As I wrote in the comment we need to improve the support of subtasks to reduce motivation to solve easy versions separately.