How are Caffine Marks measured?

[Joaquin]

How exactly Caffine Marks measured?

If Caffine marks merely cover the interpertation to real machine code, 
then the Caffine marks will not represent the true performance.

For Java applets that use AWT (or IFC, JFK, AFC) the performance can be
slower because the low level video class called peers will in turn call
the Native OS graphic libraries.

In the case of Windows '96, the GDI libraries are STILL 16bit (see
http://www.austin.ibm.com/pspinfo/chg8.html). Other environments may yield
slow performance, e.g. Motif vs Xlib, Xlib vs Berlin, etc.  The graphic
statistics need to be measured as well to help determine the true
performance.  [Psst, someonne could invent Mocha marks?]

Some more questions:
  Is there info out there on how JavaVM's compare?
  Are there tools w/ source to measure this?

I'm going to check out the Sun's website to see if I can find this.  A
CaffineMark tool could help profile algorithms.

 - Joaquin