In the end though, I do expect that the hassle of setting up and managing a solution like this is likely to result in most people deciding that it isn't worth the extra complexity just to avoid upgrading a low fib device where a larger one is really needed.
Quite the contrary, nearly 10 years ago (just before SIR was released) I was doing this precisely because the lower fib box was a good tradeoff between ports and cost, and needed something to do IXP/PNI peering with. Only instead of running the sflow analysis on the box, I was exporting it elsewhere and pushing prefix filters every once in a while to make sure the highest traffic prefixes were served locally.
Ultimately, it's part of the TCO of your network, and when traffic volumes are high, you look for any opportunity to reduce that CapEx cost of a fully high FIB router.
It sounds like the idea still has value!
M