I think it might be true for CPUID/stepping, but not for what are usually considered to be the generations.
For example, there look to be at least 5 dies sizes for Sandy Bridge: http://en.wikipedia.org/wiki/Sandy_Bridge#Models_and_steppin...
I haven't tried to match up the supported ISA's with the different die sizes, but I presume they'd correlate. Do they not?
The server designs (LGA2011) have QPI links and may be the only memory controllers capable of using registered memory. But among the processors targeting the LGA1155 and mobile platforms, they only had one core design and only one memory controller design that was shared across all products though never with everything enabled at once.
None of these features are unavailable due to being physically absent. They're just turned off. If Intel's microcode format and DRM were reverse engineered, some of the features could be turned back on.
That is honestly probably more of a factor than anything for the segregation between models. Get that yield up to improve margins significantly.
The answer is that you only save power by not doing at all the things those features accelerate. It's never more efficient to implement AES through general-purpose instructions than using the fixed-function hardware. If you've got a workload that can benefit from the wider SIMD units, it won't save any energy to only use the narrower instructions. If you're trying to secure or virtualize your OS, doing it with VT-x and VT-d will be so much faster than emulating it in software that even if those features added several watts to the chips power consumption you would still save energy.
Doing less can save power, but doing the same amount of stuff the hard way doesn't help anything.