java bytecode considered bad

[Richard Kenner]

    BTW, Java bytecode isn't really a good example of an IR because it is
    semantically much closer to source form (in fact gcj generates bytecode
    directly from syntax trees) and most interesting optimizations (e.g. 
    method inlining) cannot be performed without breaking verification.  It is
    probably better described as a form of obfuscated source.

To me, I think this is the key point, and why I disagree with RMS.

Although I understand his concern in general, I simply don't think that such
an approach is going to yield acceptably-efficient code and *certainly*
would not be a path that would allow one to take advantage of GCC's 
optimization with proprietary front-ends or to use proprietary optimizers
with GCC's front ends except in very narrow circumstances.

I see JBC as an *interoperability* issue, not an efficienty one.  Quite the
opposite, actually.