We can do that precisely for fluorescence and many other traits with GM today, but good luck trying to breed a fluorescent horse, it could take millions of years if at all.
Besides, GFP fluoresence is the proverbial low hanging fruit as it is robustly expressed in many cell types but has very little practical value. Inserting useful traits via GM is often no short cut because they are rarely determined by a single gene.
Your point about multiple genes is a very good one but we have a good understanding of how some gene networks work already and the rest will follow.
I'd say that our ability to interact with the genome is still rather crude, else we'd have already cured cancer by now - we already have a pretty good understanding of the cellular biology of cancer but it only helps us so much.
On the other hand, one can slowly breed for desirable traits such as speed, endurance or disease resistance in horses, good luck doing that with GM.