Good day rjk_cmh,
Indeed, Ethernet has a lot of benefits and is extremely well supported by a huge number of companies that provide a lot of equipment. However, pushing Ethernet all the way down the smallest controller/sensor level I can see a number of issues:
1. Home-run (back to a switch ... etc) wiring
One of the larger project costs is the labor/effort to install wiring and/or conduit. If one has 100 or so controllers and each one needs to be wired back to a switch, then I can see this cost going through the roof and potentially a lot of space (conduit), as in many cases all wiring has to be placed in conduit. Secondly with the price of wiring these days, the physical amount of wiring needed would also be quite a bit larger... which also adds to cost.
Daisy-chaining minimizes this overall installation cost dramatically.
Using wireless like Wifi is an option, however, at the Wifi frequencies uses there are limitations with regards to signal penetration and integrity in the areas where a lot of controls are installed... The higher the frequency the harder it is for the signal to penetrate concrete, stone, etc... unless bumps up the transmitted power... but the problem here is that this is in the Microwave region and too much power results in some serious issues with regards to living creatures.
2. Communication length limitations.
As you stated, Ethernet is technically limited to 100m unless one uses Fibre or Wireless, etc, whereas RS485 is spec'd at 4000 feet (assuming the data rate is < 100Kbaud).
3. Ethernet adds a reasonable amount of $$$ to a product. There are a number of components that are necessary for a wired Ethernet interface (Magnetics, physical layer interface, etc) as well as the additional resources (memory, clock speed, etc) needed by the controller's processor... translation... bigger and more featured processors costs more $$$. This also adds to higher assembly and testing costs, higher support costs, but also the increased effort and cost of the operating firmware (Ethernet require some form of a protocol stack, etc). Another cost is the increased Governmental testing cost, as the bigger/faster processor means that the product will most likely radiate more electrical emissions and so add more costs to deal with all this.
4. Adding wireless Ethernet is an option... but this tends to be costly from both a component and testing perspective, but also time to implement. Governmental radiated emission testing on devices that actually transmit wireless information (i.e. intentional radiators) increases exponentially... As an example radiated emission testing is around $5K a pop... intentional radiator testing... around $50K... a pop...
5. Reliability
Ethernet can be less reliable, as the physical layer interface (PHY) tends to consume a fair amount of power and the offshoot of this is heat dissipation. Statistically the cooler a semiconductor is operated, the longer it will live... In fact if you reduce a semi's temp by 30C its lifetime will increase 10 times (statistically). Since Ethernet interfaces are power intensive they tend to not live as long as interfaces that generate less heat (i.e. serial, etc). Again this is just statistics and so your mileage may vary.
Another point is that with Ethernet there would be a need for more secondary components like switches, etc. Again statistics tell us that more components lowers the reliability of the system.
Anyway, I am sure that others will comment, but these are few of the items I see.
Cheers,
Sam