There is no direct correlation between difficulty and price, but difficulty follows price.
Difficulty is set by hashing power. The network attempts to maintain an average time between blocks of 10 minutes. If hahsing power rises (people deploying new miners) then avg time will fall. The network correct that by raising difficulty and thus the new hashing power isn't any faster (combined) at finding blocks then it was before.
Say difficulty is 10 TH/s. As an example if 10% more hashing power comes online (~1.0TH/s, or 11 TH/s total), time between blocks at current difficulty falls to ~9 minutes. At the next difficulty adjustment difficulty will rise 10% and the 11 TH/s network will find blocks in 10 minutes (no faster than the 10TH/s network did).
So what keeps difficulty from going to elventy bazillion? Economics. GPU cost a certain amount, and use a certain amount of power. Thus price rising leads to rising difficulty (as new miners chase higher profits) and price falling leads to lower difficulty (as marginal miners quit). Trying to predict difficult OR price is a fools errand but the relationship between price AND difficulty is relatively stable:
Over last 90 days 1 GH/s of hashing power has earned between $3.00 and $6.00 per day (minus electrical costs).
When price/difficulty gets too low (either due to rising difficulty or falling price) it becomes unprofitable and some miners quit raising price/difficulty. When it gets too high it encourages new hardware spending and price/difficulty falls.
BadBear was indicating that over the long run FPGA change that. They have high upfront cost but low operating cost. Thus once bought they can remain online and profitable even when price/difficulty falls to levels that would idle a GPU. Since they don't idle that is going to put downward pressure on price/difficulty. As more and more FPGA are deployed one would expect the price/difficulty ratio to fall (network getting mrore efficient).