The oo?.p? oo benchmarks are as follows ('+' = has, '-' = has not,
'?' = kind of has):

		pasm	pl	py	rb
oo1	new1e5	+	+	+	+
oo2	new5e5	+	+	+	+
oo3	get	+	+	?	?
oo4	set	+	+	?	?
oo5	get	-	+	+	+
oo6	set	-	+	+	+

The oo1-oo4 test creation of objects, the oo1 creating 1e5 objects
and the oo2 creating 5e5 objects.
The oo3 and oo5 test getters, 2 times 5e5 gets to a single object.
The oo4 and oo6 test setters, 2 times 5e5 sets to a single object.
The oo3 and oo4 are speed cheats for Perl because they directly
manipulate the blessed reference.  No such cheats are used (or available?)
for Python or Ruby objects: therefore for Python and Ruby the oo3 and oo5
are identical, as are the oo4 and oo6.  For Parrot the oo3 and oo4 cheat
similarly to Perl since object attributes are currently implemented
only by offset, not by name. In the oo5 and oo6 the getters and setters
go through "real" (named) method calls even in Perl, but not yet for Parrot.



