XEmacs had things (multiple terminals attached at once, internationalization, color themes, input methods, and so on) long before (like a decade before) GNU Emacs had them.
But now GNU Emacs has them, so the only people still using XEmacs are approximately the people who got set up on it 20 years ago and don't feel like changing.
It still exists because things don't pop out of existence when their main reasons for having been created long ago go away.
I've used XEmacs all day every day ever since it was Lucid Emacs. I probably could switch to GNU Emacs if I really wanted, but I'm used to XEmacs, and I have a lot of personal customizations that would have to be changed. I don't really know how hard that would be, but I'm very glad that XEmacs continues to be maintained nonetheless.
Having used XEmacs for a very long time, I recently tried switching to GNU Emacs just to gain access to one particular Emacs-only feature that seemed supremely useful to me. I spent a couple of days hacking my .emacs file to get GNU Emacs configured more less the same as my XEmacs config, but even then things weren't quite the same. In the end, the "one feature" proved not to be as useful as I thought it would be, and I went back to XEmacs. Reason: call it a combination of my old habits and the silly but nigglingly painful differences between the two.
The strangest things here to me are RMS's allergy to data abstraction (insisting on using bare Lisp objects for things like events) and his excessive parsimony in design (unifying keymaps and menubars, when the only thing they have in common is that they're both trees). Neither of these reflects the software engineering practices that were taught to most of us at MIT. Perhaps he picked it up from hanging around some of the older hackers who were used to working with extremely limited machine resources.
Certainly in its early days, GNU Emacs was criticized for excessive resource consumption by people who were trying to run it on relatively small machines ("Eight Megabytes And Constant Swapping" was one joke I recall). Perhaps RMS was influenced by these criticisms; and probably, the Lucid developers did not feel quite so constrained, given their market. In any case, it's been a very long time since 8MB was a lot. It's funny to me now, when I occasionally use one of the fancy IDEs like IntelliJ IDEA or Eclipse, to see how XEmacs is now the lightweight, fast choice by comparison. (And it loads in a 500MB log file in a couple of seconds -- try that with your IDE.)
The original question is why it still exists. The answer is simply that XEmacs still has users. I would guess that only the tiniest sliver are "new" users. Some of the existing users are surely developers who continue to maintain the project, which at this point should only be of historical interest to potential new emacs users.
The same way you could ask why GNU Emacs still exists. Because there are people who for various tiny (rational or irrational) reasons dont want to switch the editor they've been using the last 20 years.
It is the same thing as asking why German is still used, now that English is a world language.
But now GNU Emacs has them, so the only people still using XEmacs are approximately the people who got set up on it 20 years ago and don't feel like changing.
It still exists because things don't pop out of existence when their main reasons for having been created long ago go away.