I imagine it's that, but amplified by NIH syndrome. Something like "oh yeah we could do it with XMPP, but it's all complicated and stuff and if we just built our own protocol we'd be able to do everything we'd ever want exactly how we want it! It'll be the future!"

I've fallen victim to something similar myself many times. "Oh yeah, I could do <this> with <tool x>, but it's ugly and not how <tool x> is meant to be used! I'll just design and implement <tool y> myself to do exactly <this> and it'll be the future!" It's the temptation to make something customized for yourself and it's hard to resist. Particularly when the existing solution has some flaws, which makes the relative cost/effort of rolling your own solution seem lower (even when it inevitably turns out to still be quite high).