Yes, that is usually the challenge. Especially a lot of repeats of the same dummy data.
Perhaps an idea would be to have a framework, and depending on what data needs to be tested, have different, customizable scripts for it.
So, if you want to test addresses, have a script that generates dummy, but realistic addresses, and so on. I'm thinking a "dummy data" generation database of its own that would house lookup tables to serve as the basis for the different types of dummy data that is needed. I'd think that for most cases, a set of common data would suffice (names, phones, emails, addresses). But the basic idea could be tailored for different uses.
Not the whole answer, I know, but maybe it's a start.
A SQL query walks into a bar and sees two tables. He walks up to them and says Can I join you?