In that time, nVidia and AMD have only really had one new generation each, and TSMC has had only one new process. Fermi to Kepler, and VLIW4 to GCN, were fair improvements, along with a minor improvement from 40nm to 28nm. There's only so much they can get do within $200 MSRP and <200W TDP. As such, they are trailing what Moore's Law should simplistically give you for performance improvements (which actually works out pretty well for things like GPUs).
Like Intel with their CPUs, one of the changes of the last several years has been that they try to game their production numbers perfectly, so when new chips come out, they don't have too many old ones hanging around. So, the old chips don't go down much in price, but simply
go away. Instead of old tech being cheaper, and having a 1-2 year tail, they just offer new cheaper tech (the shrink&rename cycle), so the last gen can get one short sale period to clear out inventory, if it's being replaced. They can't always do it, but that's what they are going for.
Well, I'd say the price range of "mid-range" has moved. If high-end is $1000, mid-range must be more than $180. But, yeah, video cards haven't been a real bargain lately.
Midrange video cards have been around $200
+50 since the Voodoo days.