After many many hours of looking at evil intel code and various
internet source we discovered that Mac Solo CPUs are quite sensitive
to instruction alignment issues.
By tweaking the compiler a little, ignoring the obsolete gnu register
overrides, plus providing march and tune commands more fitting to
the Solo this results in a VM that is 20% faster for bytecodes and
15% faster for Sends than the earlier shipped 3.8.11b2U version.
Beyond that there are also some minor changes:
a) Use PageSize boundary for allocating start of Object memory heap
(likely never used).
b) Alter getSystemAttributes: 0 to return the full path to the
executable, versus the path to the application.