I'm not saying there's no room for improvement, but you're basically describing the fundamental problem.

Higher quality audio tends to take up more data bandwidth in the wireless protocol, and resilience against interference (and retransmission or error correcting redundancy) will require a longer delay between receiving that signal and actually playing that signal. Some codecs make use of much more efficient ways of turning high quality audio into a lower bandwidth signal, but those usually come at the cost of computational complexity in encoding and decoding - which sacrifices the size and battery life of the wireless device decoding those signals. Or, some codecs allow for more efficient encoding or better error correction, but need to operate on bigger chunks of audio at a time, which might mean that the codec waits for an entire chunk to finish before it gets encoded and sent, which means that latency at a minimum is the length of the chunk. As a result, wireless audio transmission generally needs to trade between audio quality and latency.

With keyboard and mouse data, it's very, very simple. There are only so many possible keys/buttons, and even the mouse movement is essentially a two dimensional vector with an x-axis and a y-axis in the fixed amount of sampled time. That means less compression necessary to fit the data into very tiny slivers of time, that allows for the polling/refresh rate to be really high, and therefore communicate in a low latency manner.