I think there are some bsic misunderstandings about the PXA270, the PCA Generation 3 chipset, and XScale. If I am wrong anywhere at all, please correct me.
Basically, a hot topic of optimization is the so-called Wireless MMX. These instructions are simply instruction fragments that operate on low-quality, high-compression graphics and audio input streams. Judging from Intel's website, these aren't very similar to the MMX found on Pentium 2 and latter processors. The emphasis here is on not similar to: this instruction set focuses mainly on decompression and resampling. This helps audio (but much less significantly so when we look at things like MP3 encodings) but doesn't help video. The processor does't have any improvements that improve blitting onto the framebuffer. It especially doesn't have an instruction set dedicated to handling graphics transformations like rotation and scaling.
Something like VisualBoyAdvance, an emulator that doesn't perform too well, stands only to gain from clockspeed and memoryspeed improvements. Recompling it with the latest GCC might help a little, but generally speaking XScale doesn't differ significantly from the ARM instruction set in any way particularly helpful to games. Additionally, Intel provides only a Windows CE and Non-User-Space optimizing compilers, meaning the Linux kernel can be recompiled with their far superior compiler, but Linux programs canot.
Most importantly, Intel's new graphics coprocessor is designed to fill that gap. It isn't in their market interest to integrate dedicated graphics hardware into the processor or the chipset. Realistically, they have no competition right now in terms of graphics processors on the PCA chipset, because no one else makes them. The Imageon series, as of now, isn't forwards-compatible.
If there are any revisions-statistics are welcome-please post them.