Yet another test on the inter-webs that doesn't show or mean anything. I would gladly challenge your composite results and overall conclusions, because essentially all you've done was to take the easy way out of such testing.
Those browser, or anything for that matter, would be properly tested in a scenario that involves first of all both 32 bit and 64 bit versions of the browser on the native OS kernels. Second you should test all browsers (if available) on all systems:
winXP, winVista, win7 32 bit and 64 bit
MacOS X Lion again with 32 bit and 64 bit versions of the browsers
Linux - 2.4, 2.6, 3.0 32 bit and 64 bit. Again on 32 bit and 64 bit versions of the OS, not a multilib-ed Linux. And here you should probably stick to some Linux OS that has the stability in mind, lets say RHEL and Ubuntu-LTS or Debian Stable
Also you should run the tests on all BSD variants and on Solaris or OpenIndiana or whatever rocks your boat.
Perfect scenario would be to have DTrace running on Solaris, MacOS X, Net|FreeBSD while doing the tests once then again do them without debugging. Also throw in SystemTap for Linux also. And the appropriate (if applicable, I don't know MS products well) windows counterpart of Dtrace/SystemTap.
Also all test should be run at least 7 times, with the odd result taken out and the rest of 6 forming a composite number.
Then you should publish all your results in a downloadable form, so everyone can see them and do their own conclusions, might be most of the conclusions are like your, might be not. Either way, that is a proper scientific method. Everything else you've shown here is essentially rectum-stats. Provided you follow such a method, you would actually work for a living, because you would need at least a month or so to complete it properly. After all everybody should work, if you are short handed I hear the people looking for jobs in the States are quite a lot, I bet you could find at least one person happy to join in and do a proper job for you.