I think that for the most part obfuscation is pointless: even with full source code it's generally hard enough to figure out what the heck intention was (assuming there are no comments, and no meaningful names for local variables -- which is the case when re-generating sources from byte code). Obfuscation just decorates the cake.
I think developers and especially their managers tend to greatly over-exaggerate risk of someone seeing the source code. While good decompilers can produce nice looking source code, it's not trivial to work with it, and costs associated (not to mention legal risks) are high enough to make this approach seldom useful. I have only decompiled to debug problems with closed-source vendors' products (deadlocks in DB abstraction layer, ugh).
Bytecode was actually obfuscated, I think, but we nonetheless found the underlying problem -- it was an actual design problem.
