OK - for starters, the OMAP36x0 (X), Snapdragon 8x50 (EVO) and Hummingbird S5PC100 (Galaxy-class) are all variants of ARM Cortex A8 cores. References provided upon request.
Not that that really matters - the Android native development kit - that is to say, the part of the tools that lets developers fine-tune for highest performance - kit goes as high as ARMeabi v7A with NEON and VFP support (along with thumb, thumbee, edsp and fastmult). See:
Android NDK | Android Developers
Bottom line - if all you can get to are ARM v7 features with extensions - then that's all you'll see for performance.
For graphics, using the critical Mtps (millions of triangles per second) benchmark:
Hummingbird s5pc100 - 10 Mtps
Snapdragon 8x50 - 22 Mtps
Omap 3640 - usually benchmarked in the same range - 20+ Mtps
SAMSUNG's Digital World
Qualcomm Products and Services - The Snapdragon? Platform
Ruminations on various benchmarks for the OMAP 3600s, Hummingbird, and Snapdragon | AlienBabelTech
Bottom line - none of these are particularly bad GPUs.
Part of the real debate would include the DSP chips - together with the GPUs, they're responsible for video and image processing.
The Snapdragon has a higher speed DSP than the OMAP3 - but, TI claims better pipelining for their unit, so it could either way or be a toss-up.
As for benchmarks:
A good OMAP 3640 vs snapdragon vs humming bird article - xda-developers
Let's look at AnandTech again:
Motorola Droid X: Thoroughly Reviewed - AnandTech :: Your Source for Hardware Analysis and News
OK - frankly, their credibility is a little low right off because they identify the Apple's A4 chip as using an SGX535 when in reality it uses an SGX530.
They also persist in claiming that the Snapdragon uses an Adreno 200 GPU, and that's fine - except the part where Adreno seems to be the name for the integrated AMD Z430 and its accompanying DSP module. (Qualcomm did buy that AMD division, by the way.)
My problem with that particular benchmark is that so far as I recall, those same performance metrics (fps) seemed to apply to the Nexus One on 2.1 - but at 2.2, the Nexus ran just only a tiny bit faster than the Droid X - running at 2.1.
And frankly - unless something surprising happens - I think we'll end up seeing all of these processors leveling out with roughly the same performance with 2.2.
In fact, I think Moto knew this and that's why they backed down from their 2.2 release promise and went with 2.1 - everyone's simply convinced that they're going to see so much better performance with 2.2 - uh, ok, maybe.
OBTW - Anandtech was pretty clear that the Snapdragon uses a Scorpion instead of an ARM A8.
Mmmmk. Let's notice that a Scorpion IS kinda an A8... except for the part where a Scorpion uses a 128-bit parallel data path as opposed to 64 for a straight A8.
Inside DSP Articles
None of these are straight A8s - they're all pretty much equivalent variants.
The benchmarks are just showing overall performance with a given software stack.
And because the software stacks vary, we know next to nothing about the processor performance differences other than how interesting they are.
By the way - just one more beeyotch slap against Anandtech - the Snapdragon 8250 and the 8650 are 100% identical - except for which 3G radio chipset they talk to. So - that should prove completely that they didn't benchmark GPUs like they said.
They benchmarked phones and talked about GPUs - instead of their software stacks.
Useless.
We get to do it again with the Hummingbird S5VP210, Snapdragon 8672 and the OMAP4 processors when those are fielded.