When new Mac are offered, anyone wants to compare them to old ones, or other Mac Family. But are these benchmarks fair? Are you sure these tests doesn’t compare Apples to Oranges? How do you retrieve benchmarks results in your day to day work with your Mac (or your games!) ?
Fair benchmarks?
They are Fair, but they didn’t test the same thing. GeekBench measure multi-threaded CPU cores in Integer and Floating-Point, and RAM bandwidth, ignoring CPU-cache size and speed in any way. XBench try to give a view of balanced speed of Mac subparts, balanced as a PowerMac G4 where CPU was the limiting factor and hard-drive relatively fast compard to PowerPC G4. MacWorld SpeedMark 6 gives an overview of application-level performance (real-world speed for these applications).
So they doesn’t measure the same thing, GeekBench is sensible to number of core, cpu frequency and DDR3 instead DDR2. XBench is sensible to balanced configuration, where CPU, Memory, OpenGL videocard, and moreover hard-drive subsystem is scaled up from PowerMac G4 balance (a good one from my point-of-view). MacWorld SpeedMark 6 will tell you which level of performance to expect on some applications and some selected tasks.
So they are fair, but you have to digg to understand what their Results means…
Benchmarks Results in real world?
Geek Bench overall score correspond to heavily multi-threaded optimized applications (kinda Pro applications) such as CineBench or H.264 encoding with x264 (FFMPEG, handbrake and many other applications). It’s memory subpart doesn’t really correspond to any application as influence is really below the cache size and performance that is not tested. Integer and Floating point performance will give you an insight for these kind of applications, for me the Integer test is the most revealing of raw cpu performance-level in many case.
XBench will reveal a balanced configuration, for example putting 2×1TB 7200 rpm in RAID-0 in a first-generation Mac Pro instead of the crappy 320GB hard-drive will enable it to go from 115 to 170+, Cinebench won’t be faster or game, but in day-to-day use you will find your Mac much more responsive, application launching faster. It doesn’t reveal a faster CPU on new iMac because with Core i7 iMac you still have a single 1TB 7200 rpm hard-drive, marginally faster in real-world performance to what you had on iMac G5 for example, while CPU is 8X faster! Launch an heavy application on both and you will discover that hard-drive is a limitation for today’s Mac, except upgraded Mac Pro 
MacWorld SpeedMark 6 will give you a good view on the performance-level for specific tasks on some applications. But if you use another application or another task in the tested applications, it will be of no help to know if your Mac will be faster or slower than another. Radeon 4670 shine one Call of Duty, but it’s a dog when using OpenCL applications and you might prefer a GT120 for that matter. You may use mono-threaded “compressor” for HD Video, but if you use heavily multi-threaded x264 based application, results will vary! Aperture is used for photography but the big majority of photographer use Adobe LightRoom 2 that is really different in it’s internal conception and multi-tasking use or GPU use!
All-in-one, none of these benchmarks will give you a complete point-of-view about your Mac with your Application, except SpeedMark 6 if you use same Applications and tasks.
How to use these benchmarks?
You will end-up looking at benchmark global score, where you have to take a look at sub-scores: these are much more revealing. You must compare sub-score between Mac to see how they will fit for your use, for example…
Gamer: OpenGL results of XBench are totally out (old 3D), you must focus on SpeedMark 6 Call Of Duty performance-level that is most revealing. Forget GeekBench!
User of heavily multi-threaded application, with few disk IO (ie. x264 video encoding): Focus yourself on the GeekBench Floating Point score, SpeedMark 6 CineBench, Handbrake and MathematicaMark scores. Forget XBench!
Apache/Mysql/PHP-Ruby developper: You must focus on GeekBench Integer Score there’s no floating point involved, XBench is of no-use (until you need heavy disk IO!), SpeedMark 6 too.
Launching many applications at-once (after boot) and waiting: look at your XBench disk score, and try with faster hard-drive, or with Mac Pro install 2 or 3 new hard-drive in RAID-0!
Having a mix of applications, of different kind, try to find the kinda application you use in MacWorld SpeedMark 6 and do your own metric
To resume…
GeekBench, XBench and MacWorld SpeedMark 6 are fair tests, but they doesn’t test the same thing (subparts versus Application-level performance) and subpart test are oriented among CPU+RAM or whole computer subparts, with maximum performance or balanced performance as primary target.
None of them could be a true measure of a Mac performance, because the performance of YOUR MAC depends on the Application you use and the tasks you perform. It’s something personal and none of them could be a true metrics that you may definitely apply to a Mac.
But you could do you own Mix to see wether a Mac will be faster or not for your own use, remembering that 20% faster is hardly noticeable, and 40% faster on task execution may justify a move on long tasks (but not short ones), and you need 70% faster to see differences on fast actions!
So upgrading from 2.8Ghz to 3.06Ghz iMac won’t give you any real juice, albeit upgrading from Core2 Duo 3.06Ghz to Core i5 2.66Ghz will be impressive
