Registers hold only a handful of values while main memory holds billions. Why does the CPU keep so tiny a store at all, rather than working directly from memory?

A. Because registers are readable within one clock cycle and memory takes hundreds.

Registers sit a fraction of a millimetre from the ALU and read in one cycle; main memory is centimetres away and takes about a hundred. The smallness is not a limitation to overcome but the reason it is fast, more registers means longer wires and larger multiplexers. They are the most expensive storage per byte in the machine, and few because being few keeps them quick.