We try to implement everything as accurately and as close to the original Cafe OS/hardware as possible. But since large parts of the the Wii U architecture are layers of abstraction it can become hard to define what 'accurate' means. For example, there are different versions of the OS libraries (based on installed console patch) and they often exhibit different behavior in some functions. As such, there are multiple solutions to how we can implement them and all are equally 'accurate'. Overall, it's a rather complex topic and the initial question cannot be answered without writing a 1000 page post that analyzes every part of the emulator. It's also important to note that the biggest source of inaccuracy in CEMU currently comes from unimplemented features.
We certainly do not rush and if something is implemented inaccurately it's either by accident or because we plan to improve later. Due to the complexity it's sometimes not feasible to implement everything perfectly on the first try. That's why we sometimes go back and rewrite large parts of the emulator, like the shader decompiler (in v1.1.0) or our
FS code.
On a side node, both Decaf and CEMU use the same HLE approach for OS library emulation which abstracts some of the direct hardware access away. This doesn't decrease accuracy from the point of software running in the emulator but it can make a difference when trying to do anything 'outside' the CafeOS, e.g. accessing the kernel via exploits. A true low-level Wii U emulator that is 100% accurate down to the lowest level will likely never exist because it's an incomprehensible amount of work (we tried initially, but soon realized it's 100x times more work at almost zero benefit).
You have to realize that looking at just one or two games is not representative of the state of the emulator. We are trying to emulate a platform with hundreds of games. With every change we make, no matter if good or bad, something will break somewhere. Even if we improve one part of the emulator to be 100% perfect it can still trigger a chain reaction that causes games to stop working until we implement some other feature. In short, there is no guarantee that any game will work unless CEMU is 100% complete, even if the very same game worked in all previous releases.
In regards to performance: Emulation of more features decreases speed, obviously. There is still lots of room for performance optimizations in CEMU, but it's currently not a priority.