It's interesting how software got better over time on the same hardware back then.
My first was a 48KB Sinclair Spectrum which I used for at least 5 years, and the games coming out by the end of that time were much more sophisticated than the early ones, e.g. filled 3D polygon graphics over blocky 2D sprites. The programmers just had to squeeze more and more out of the same hardware.
These days, a similar jump in sophistication just doesn't happen without new hardware.
To be fair you do see the same phenomenon on game-consoles. The hardware stays the same, but newer games are always pushing the limits just a little bit further.
I actually think hardware stability, predictability and closeness is central for this to work.
Once a platform is new, nobody really knows what it can do and how it can best be exploited. Like on the Commodre 64, you had a limit of 8 hardware sprites, each one either 32x32 pixels mono-coloured or 16x32 pixels tri-coloured (iirc). This didn't really allow you to do much fancy games or graphics.
But once people discovered that you could reuse those same sprites multiple times during a single screen-refresh by timing your sprite-programming to whatever scan-line the machine was outputting, the genuine was out of the box: The Commodore 64 and its games would never be the same again.
With changing hardware or a hardware abstraction layer, this would probably never have been possible to do reliably or at all. I'm willing to guess that having hardware stability, predictability and closeness can be listed as a requirement for allowing developers and games over time to discover just how far something can be stretched.
For phones and a million other different mobile appliances I think this will be much harder, although not impossible. I doubt will see the same sort of mad capability leaps over time on these devices. But hey, you never know ;)
