Question about graphic cards: Memory Clock vs Memory Bus?

1. For customers there's no difference. What matters is the final number: multiplication of bus width and frequency, that'll give you the transfer rate. Higher memory width is more expensive to manufacture (each bit width is an actual physical lane and is the total amount of pins in the memory chips dedicated to transfer data), and higher memory clock is harder to achieve.

2. The frequency determines how many transfers can be done per second (2 per cycle, actually), and the bus width how many bits are transferred simultaneously.

3. Power requirements and heat are important limiting factors. Also, having too much transfer rate becomes useless. Just look at AMD cards with HBM2, they have much higher transfer rates but don't really perform better than the equivalent nVidia with GDDR5. In other words: it's useless transferring more data than the GPU is going to be able to process.

edit:
I'm going to add something that I'm not sure to what extend is a problem in this case: transferring data over parallel lanes creates interference due to electromagnetic induction (this is why ethernet cables are twisted, to minimize interference). So if you have many parallel lanes sending and receiving data, there are interferences affecting each other. This interference increase as you increase the frecuency of the signal, so in theory, it's harder to have higher clocks with more lanes.