Synchronized methods generated by compile_to_java

I'm aware that compile_to_java is not working yet.

I was wandering how the compiler will decide for which routines to
generate synchronized policies.

Three possible strategies spring immediately to my mind:

1) No routines are synchronized.
2) All commands are synchronized but no queries are.
3) All routines are synchromized.

and I expect more complex policies could be devised.

The reason I'm speculating, is that I have been observing that Saxon
(an XSLT processor written in Java) gets a free performance boost when
running on an SMP machine, in as much as the JVM is apparently
parallelising execution in some way - it uses more than one CPU, yet
there is no multi-threading present in the Saxon code.

I presume that the JVM will not parallelise execution of synchronised
routines, which lead me to think of this subject.

Does Smarteiffel support any multi-threading of any kind? As far as I
know, it doesn't, in which case (1) might reasonably be selected as a
valid strategy. But if the JVM is then going to multi-thread under the
covers, it might be disastrous.
(2) looks safe, but then I have to worry about memo functions.
(3) is certainly safe, but it would be a pity to loose a free
performance boost.

Am I making any sense?
-- 
Colin Adams
Preston Lancashire