I must admit, i did not find out how exactly to register.
I understood that some type conversion operation
shall be registered.
If you tell me the signature and how to register,
I can fill in the implementation.

-- or did I misunderstand something??

Hello Mike,
I would like to provide a patch for this.

I would write an m-file double.m
delegating to the builtin function double
except for java types.

For those I would check for java.lang.Number
and then apply Number.doubleValue().

Is this acceptable??

http://de.mathworks.com/help/matlab/matlab_external/handling-data-returned-from-a-java-method.html

shows that in Matlab function double may be applied to java objects.

For Octave:

bd=javaObject('java.math.BigDecimal',42)
bd =

<Java object: java.math.BigDecimal>

octave:26> bd.doubleValue()
ans =  42
octave:27> class(ans)
ans = double
octave:28> double(bd)
error: invalid conversion from octave_java to scalar

this should not be.
In fact, double(bd) should invoke bd.doubleValue()
as it is an instance of java.lang.Number.