A large part of why this is hard is they tend to produce something other than what you want. Getting a protean to fold the right way 95% of the time is a huge achievement, but it also means your producing something else 5% of the time. In point of fact quite a bit of your DNA is focused on getting other sections to fold correctly.
As to population risks Prion's are probably a risk. http://en.wikipedia.org/wiki/Prion For one thing there unlikely to be undetected in the food supply as normal toxicity tests will miss a novel one. (EX: Mad Cow Disease)
Anyway, all I am really just suggesting GMO’s are introduced more slowly. Something like food for 100k people in the first year, 1M the next, then 10M and then you’re good to go. Granted, that would be really hard to do with the way the current food supply works, but simple acreage limits could have a similar effect.
PS: The point about Hedysarum alpinum toxicity was it's not all that uncommon and people had been eating it, but it was not known to be toxic. The point being Toxicity can be more complex issue than is generally assumed.