In fact, they did what?

The world is full of code-compatible processors where the successors have added extra instructions. Sometimes with full binary backwards compatibility. Sometimes with assembly-level compatibility. The important thing here is that 8 bits represented a very decent addressing range for a microcontroller the 8051 was designed. And the 16-bit extensions added allowed access to truly tremendous amounts of memory for a microcontroller.

My historic note about the PC aimed to show that the design of the 8051 allowed it to address as much memory as a fully populated motherboard could swallow of the initial PC. 64kB was HUGE, so there really was no real need for Intel to add more 16-bit addressing modes. Few people could afford a PC with 64kB of RAM, and even fewer where likely to be able to use a microcontroller with such vast amounts.

In short: Intel did what was needed to access a 16-bit range, but had no real reason to go one step further and make the 16-bit addressing modes reasonably orthogonal. You don't spend extra transistors on features you don't believe your end users are likely to regularly use.