1. interaction peripherals where Bluetooth+batteries is workable (keyboards, mice)
2. interaction peripherals where Bluetooth+an external power cable is workable (speakers)
3. low-power data peripherals (CF cards, flash drives) where low-speed USB is acceptable
4. high-speed data peripherals and controllers (RAID storage, discrete GPUs, etc.) that effectively need a PCIe channel
1 and 2 are wireless, obviously. 4 doesn't get reconfigured much, so it makes sense that Thunderbolt and related "extension" ports are on the back of the device. (They were previously inside the device, after all.)
3 is the main problem, then, that currently suggests accessibility of ports: the small, transportable, label-able data-containing devices that descend from floppy disks.
But why not think of alternative solutions, if "lightweight" data storage devices are the only constraint?
One design I always thought fondly of was having small magnetic-backed data-storage "pucks" with no plugs as such, and making the non-screen part of the computer (e.g. the front-bottom panel area of an iMac) a magnetic attach-point for them. You would literally "mount" the your drives onto the computer. (Presumably it would inductively power them while they're there, too.)
This wouldn't be a very fast storage device class compared to something using actual wires, but it wouldn't need to be; fast storage is what Thunderbolt et al are for.
It'd likely also have other uses: for example, magnetically attaching your ID card (which is also a smart card) to a corporate computer to cause it to log in—and also to indicate to anyone else walking by that the computer is in use.
I'm not sure putting the ports on the rear of the e.g. iMac improves usability. It certainly looks nicer, but I wouldn't consider the required neck acrobatics to plug in a thumb drive a usability improvement. (But I guess it makes sense in the long-term with everything going wireless.)