1. It's a pain, both in terms of what it provides and what it doesn't provide, compared to every alternative I've used
2. Despite being an ISO standard, it's a different pain everywhere
3. Even libraries which exist only to make it more consistent and less painful don't help beyond fairly trivial tasks on the 1 or 2 most popular free implementations
If any other language had this property, I would not use it; SQL is no different.
However, two points. First, the diffs aren't that huge (the underlying idea is generally the same, with some caveats, it's just the syntax that must be juggled). Second, most places I know usually pick one or two database systems and then make a very long commitment to them. There is no reason the developer shouldn't be able to pick up whatever SQL flavor his shop is using.
My original point is that any developer worth his salt should be able to pick up new languages fairly quickly, and since SQL is like a mini-language, it should be learnable in mini-fairly-quickly time.
I'm a bit scared of SQL because it's a pain to use, and I've used object databases with great success (and no pain). For complex queries, its pseudo-English syntax is more trouble than it's worth. And one of the most common uses today is full-text search, but SQL doesn't let you search that. (Officially, it doesn't even provide INDEXes; that's probably why CREATE INDEX syntax is not the same anywhere.) I don't understand why so many otherwise-great systems (like Rails) force you to use SQL.
To take another example of a standard implemented by different companies, I've taken nontrivial Common Lisp programs from one compiler to another, and almost never had problems. I would have thought that CL would be a much harder standard to get consistent than SQL, but apparently it's not.