I understand that a big part of Sun's value was in Java, but the many-core high-I/O-bandwidth server business was still valuable to Oracle and synergized pretty well with Oracle's business (outside of the legal department). The value to the world as a whole was much higher with IBM (or, at a second choice, Google) owning Java and Oracle (or some other high-I/O bandwidth enterprise product vendor) owning the Sun hardware business.

The Jikes compiler's startup time was so much faster than Sun's javac, and Java compilers perform very few optimizations when compiling source to bytecode. (My understanding is that the spec restricts which optimizations are allowed.)

JikesRVM (a JVM written in Java) had surprisingly good performance for a small research project. It's bootstrap process involved AoT-compiling itself, so it would have been not too big a leap to have a general non-GCJ AoT for Java years earlier, and (because it uses its normal second-tier JIT for AoT compilation) presumably not too difficult to get the hot paths all re-inlined and re-optimized even through the AoT-compiled code.

I think a very small minority of developers or users preferred the look of Swing (or worse, AWT) widgets to SWT. Sure, SWT widgets looked different across platforms, but the alien look of Swing/AWT widgets was very off-putting to users (and to a lesser extent, I tihnk, developers).