That's language prudishness, right there.
I also find your assertion that if someone learns PHP as a first language they're likely to end up a "better" programmer than someone who learns Java as a first language absurd. There are good schools teaching Java, and smart, inquisitive people studying it. Many of the best programmers I know still use Java for pretty much everything since it allows them to get their jobs done. Some know other languages, some don't. It's a purely pragmatic choice, but to say that those people would not be valuable to a startup is crazy.
Maybe the process you use when interviewing is good enough to separate these people from others who really don't know how to program, but it seems like they're going to be starting at a significant disadvantage if you're interviewing them.
By coincidence—not by any flaw of Java or the generalist programmers who happen to know it and like it—the language that will most frequently appear on the resume of a single-language programmer is Java. This means—in the same way that given races of people get tested for given congenital diseases because they are statistically more likely to be afflicted by them—that if you're interviewing a programmer and all the previous projects listed on their resume are Java projects, it becomes more important to ascertain whether they know any other languages, because they are statistically less likely to.
This point is just a combination of the famous and well-received "programmer vs [language X] programmer" rant, and a bit of Bayesian analysis to determine which people applying for "programmer" jobs (i.e. generalist jobs, of the kind needed in startups) are most likely to actually be "[language X] programmers" instead. If you trained a spam filter on generalist-y-ness, it would flag Java as having a high correlation[1] with non-generalist-y-ness—that's all I've been trying to say.
[1] ...which is very explicitly not causation, I think I have to remind everyone at this point.