I'm trying to understand what factors determine the MTU. For instance, why 802.11 has MTU of 2304 while Ethernet MTU of 1500? Are there restrictions of the physical medium or arbitrary design decisions?
The MTU is the size of the payload allowed for the data-link protocol. That is determined when the protocol is standardized, and various criteria go into the decision on the payload size. The ethernet payload size was based on a compromise for a reasonable size of the amount of data to be sent against having a host monopolize the shared segment for a long period of time.
It is not really a physical medium limitation (although having a shared medium had something to do with the original decision of payload size). There are many implementations of ethernet that support jumbo frames, although that is non-standard. Ethernet also runs over various media at various speed, and the ethernet MTU is still 1500 octets for the various ethernet standards.
Trying to change something so basic as the MTU on an established standard creates a lot of problems. For example, the ethernet jumbo frames can cause big problems unless the use is carefully planned (every interface in the path must support the same jumbo frame MTU, and ethernet does not have fragmentation or retransmission, so frames too large for an interface are simply dropped).