R700 is atleast 2.5x more powerful than the xenos chip in the 360.
Actually much more powerful. While a basic GFLOPs comparison isn't the end all way of comparing Xenos to whatever other AMD GPU, it's a good indicator of relative performance since the Stream Processor set up didn't change so much from Xenos to R600 to R700 to even Evergreen and Northern Islands to make them all too unrelated.
Using Wiki for Stats, here are some GFLOPS comparisons for the R700 series:
Xenos @ 500 MHz - 240 GFLOPS
Radeon 4550 (RV710)....@ 600 MHz - 96 GFLOPS
Radeon 4670 (RV730XT).@ 750 MHz - 480 GFLOPS (possible candidate for Wii 2 GPU)
Radeon 4770 (RV740).....@ 750 MHz - 960 GFLOPS (likely candidate for Wii 2 GPU, most hoped for it seems as well :hmm

Radeon 4850 (RV770PRO)@ 625 MHz - 1000 GFLOPS
Radeon 4870 (RV770XT)..@ 750 MHz - 1200 GFLOPS
Radeon 4890 (RV790XT)..@ 850 MHz - 1360 GFLOPS
Personally out of the entire series, I do think RV740 (Radeon 4770) would be an excellent choice simply because it would bar none be 4x more powerful than Xenos (from purely an SP view) at 750 MHz. Even backed down to 500 MHz to save energy, it would still hit 640 GFLOPS. Also, it's a 40 nm GPU and Ninty could've had AMD work on getting down to 28 nm which would reduce power consumption by a great deal without a clock decrease. Even though math is probably a bad indicator of actual turned out die size (since all sorts of changes have to be made to make a node size reduction workable):
(137 mm^2 RV740 GPU x 28 nm) / 40 nm =
95.5 mm^2 correct? (assuming perfect ratio change with node size)
RV730 (Radeon 46xx) would make a great candidate as it would be pretty small on 28 nm
(146 mm^2 RV730 GPU x 28 nm) / 55 nm =
74.32 mm^2
My beef with using RV730 would be it's low ROP count (8 ROPs) and that would make 1080p difficult (but probably manageable) for the same graphical fidelity as the Xbox 360, especially since the 360 has it's eDRAM advantage.which get's it just about "free" 2x MSAA @ 720p. At 720p though, the Wii 2 would be clearly the superior GPU, especially with 2x the shader performance.
But as far as I'm concerned, RV740 is where it's at in regards to the R700 family. RV770 is probably too big for an easy process reduction and not worth the effort. RV740 is a good start for node reduction since it's already 40 nm for "instant installation" into devkits before it could be shipped on 28 nm for production consoles and at a good balance of capability/power consumption. According to 4770 reviews I've seen, power consumption is in the 50W area (entire graphics board) with the 512 MB of GDDR5. Adding in another 512 MB to the VRAM (for 1 GB total) + 1 GB of GDDR5 SRAM + CPU and keeping it all under 100W should be easy, especially if the CPU and GPU are on 28 nm.