This is a really hard question, and I don't think there's a good, specific general answer. A non-specific general answer is that you have to find something that works reasonably well, takes advantage of any pre-existing resources, is as easy and practical to use as possible, and that people are willing to use. That almost sounds like a description of the problem, more than an answer.
IPA might not actually be a good answer, depending on the variety of dialects and the phonology of the language in question. For example, some speakers of Hawaiian pronounce w more or less like English w, others pronounce it like English v. Do you change the writing based on who's talking? Another example is English t—for many speakers, the t in "top" is different than the t in "stop" and both are different from the t in "bottle"... but in the mind of an English speaker, they are all t. IPA can be too detailed.
You can also lose relationships between words if you spell only according to sound (again, depending on the language). English "photograph" and "photography" might be spelled more phonetically as "fotəgræf" and "fətagrəfi", but that decreases the obviousness of the relation between them. Is it worth it? It's a judgement call.
As for using the script of a/the dominant language of the region where the language is found, it depends on a lot of linguistic and cultural factors. Is the relationship between the language groups neutral or positive? Are people already generally literate in the writing system of the dominant language? Are the languages fairly closely related? Yes to any of those makes the writing system of the dominant language a better choice—though it still may not be a good choice if the answer to any of them is no.
Are there closely related languages (linguistically, not necessarily geographically) that already have a well-designed writing system that could be borrowed and adapted? For example, a lot of the Turkic languages have adopted fairly similar versions of the Latin alphabet. That way some problems only have to be solved once, and it can also be easier to read a closely related language.
Another technological issue—are input devices (keyboards—either physical or virtual, like on a phone) that cover all the needed symbols readily available? (That's another reason not to use IPA—I think it's actually not too hard to learn the subset relevant to a language you speak, but it is often really hard to type.)
Ideally, a writing system should be devised over some time, with heavy input from speakers of the language and guidance from linguists who have experience with related languages, the development (historical or practical) of writing systems, or both if available.
A gateway wiki might be a good place to experiment with a new writing system, but it could also end up generating too much inertia if good changes are proposed that require re-writing almost everything. I really don't know.
It's an interesting and difficult question you have. I hope it generates some fruitful discussion.
—Trey
PS: Also read up on spelling reform in various languages for related ideas and possible problems.
Trey Jones
Sr. Software Engineer, Search Platform
Wikimedia FoundationUTC-4 / EDT