The upgrade path for the Amiga chipset ended up being complicated. The 500 and 600 had the ECS chips, that allowed some video modes that were not tied to NTSC, at the cost of a reduced palette. The chipset also competes with the CPU for memory access and this second generation chipset addressed more memory, making the computer effectively slower ("chip RAM" was slow, "fast RAM" was the RAM outside the reach of the chipset that the CPU has exclusive access to). After ECS came AGA, which pushed the boundary further again (but, at this point, memory constraints were not so terrible).

And, of course, there was what seems like a cocaine-fueled endless sequence of management blunders that drove the company into the ground.

> this second generation chipset addressed more memory, making the computer effectively slower

ECS could address more chipmem but wasn't slower than OCS. You couldn't add fastmem in the trapdoor port, but that didn't matter much since those expansions weren't good enough to impact the speed (trapdoor fastmem was usually called slowfast). The problem was rather one of incompatiblity: some programs written in the 512+512 kbyte era simply assumed they could allocate fastmem, which typically wasn't available on the 500+ and 600.

> After ECS came AGA, which pushed the boundary further again (but, at this point, memory constraints were not so terrible).

AGA, like ECS, could address 2 megs of chipmem. However, it had higher bandwidth and was much faster than ECS.

My point is that even though the ST/e was, as you say, a simpler design in many aspects, Atari still had to equip the Falcon with a YM chip and put support for planar 15 kHz video in VIDEL to maintain backwards compatibility. They faced the same problem as Commodore: their machines were mainly home computers used for games and other software that banged the metal and people expected this to work when upgrading. They also shared a lot of the same problems when upgrading the architecture even slightly, such as with the A3000 and TT030: programs that didn't work with newer versions of TOS/DOS and programs that didn't work with 020/030.

Both platforms are expandable with things like RTG graphics cards, sound cards, CPU cards etc. (in fact I'd argue the Amiga architecture with Zorro, video slots and CPU daughterboards was designed to be vastly more expandable than the Atari) but for most users that didn't matter: if the games they wanted to play didn't work, what point was a 24-bit display that cost more than the computer itself?

Besides, even without keeping backwards compatibility, rolling your own silicon was no longer a viable option financially. Tramiel's vertical integration was a good idea in the 1980:s but the hardware market had shifted. Commodore could've made a triple-A machine but it still wouldn't have been competitive, neither in price nor in performance. The niche markets utilizing the unique features of Atari (MIDI) and Amiga (DTV) weren't large enough and a new architecture that would deprecate all or most existing software (and many peripherals) used by hobbyists would probably only serve to push the home user base towards the PC anyway.