Because the GPU is "ment" to do the work here. The CPU is a lot less powerful when it comes to computing lots of smaller calculations (which games do). So if a games uses 100% CPU and 20% GPU then the game is very badly optimized and would perform a lot better if the game was made to rely more on the GPU
Planet Coaster is a different type of game than the normal First Person Shooter, or Puzzle game. There will be more CPU usage because Planet Coaster is simulating thousands of guests, riding 25+ rides, walking on paths, buying things from shops, and many more aspects. All these that I just listed aren't in a normal game, and these aspects aren't something that is normally handled by the GPU. Displaying people and rides are done on the GPU, but calculating where the thousand people need to be and how they feel and move and interact with the rides are all done by the CPU.
GPUs are getting more general purpose, but the ability of them to be general purpose varies widely, and if Frontier programmed the game to run all the simulations (people, rides, money) off the GPU, people would be even more upset because likely the visual prettiness of the game would have to be reduced because half the GPU is calculating simulations (for instance). In reality the programming language support is all over the map for General Purpose GPU programming, so Frontier would have to have an even higher minimum hardware requirements, so they could target a smaller subset of graphics cards resulting in them being able to complete the game in a reasonable time and budget. Going even further they'd have to limit themselves to routines that are supported across all the graphics cards in that subset. Likely the language library that would be used is OpenCL 1.2. No using Cuda because that is only Nvidia. Nvidia is behind on their support of Open CL, only supporting version 1.2. Intel and AMD support one release further at OpenCL 2.0. The CPU still has to coordinate the GPU, so there will always be some useage there. There is even further a knowledge roadblock because General Purpose GPU programming is an emerging field, so there are less people that can do this special purpose programming compared to the number that can write code for the CPU.