Linaro optimized CM9 builds for VZW and GSM Galaxy Nexus

Benchmarks below.
Test CM9 Nightly 6/5/12 CM Linaro Units of Measurement
Linpack
Single Thread 43.6854 +/- 3.71% 43.6056 +/- 4.35%
Multi-Thread 74.5724 +/- 2.39 73.2464 +/- 1.59% MFLOPS
AnTuTu 5454 +/- 2.68 5565.2 +/- 1.01%
RAM 806.6666667 +/- 2.55 820.6 +/- 3.02%
CPU Integer 1395.666667 +/- 4.77 1429.2 +/- 4.64%
CPU FP 1102 +/- 6.45 1179.8 +/- 2.67%
2D Graphics 285.3333333 +/- 0.20 285 +/- 0.43%
3D Graphics +/- 1221 +/- 0.08 1221.4 +/- 0.15%
Database IO 378.3333333 +/- 12.49 402.5 +/- 6.06%
SD card Write 101 +/- 14.59 96.8 +/- 12.59%
SD card Read 164 +/- 21.26 154 +/- 16.94%
PassMark Performance Test
CPU 4163 +/- 5.68 4413.2 +/- 4.20%
Integer Math 88.66666667 +/- 4.76 90.66 +/- 1.15% Mops/sec
Floating Point Math 184.3333333 +/- 11.02 199.2 +/- 2.31% Mops/sec
Find Prime Numbers 45.4 +/- 11.48 50.36 +/- 0.72% Thousand Primes/Sec
Random String Sorting 613.3333333 +/- 0.57 684.6 +/- 5.00% Thousand Strings/Sec
Data Encryption 2.74 +/- 4.15 3.006 +/- 4.85% Mbytes/Sec
Data Compression 498.6666667 +/- 5.16 515.8 +/- 2.58% Kbytes/Sec
Memory 1241 +/- 0.98 1241 +/- 1.32%
Write 129.3333333 +/- 0.89 128.4 +/- 0.43% Mbytes/Sec
Read 847 +/- 0.85 850.6 +/- 2.31% Mbytes/Sec
Sunspider 2220 2103 msec
0xBench Linaro CM9 0xBenchmark CM9 6/5/12 0xBenchmark Units of Measurement Percent Change of Linaro CM9 to CM9 6.5.12
Linpack 43.11114487 42.20384869 Mflops/s 2.15%
============== ============== ============== ============== ==============
Scimark2
Composite: 64.19346113 52.33802373 22.65%
Fast Fourier Transform 41.4348302 42.09075043 -1.56%
Jacobi Successive Over-relaxation 148.0451013 94.01614855 57.47%
Monte Carlo integration 11.14765186 10.06129955 10.80%
Sparse matrix multiply 40.51434074 40.03910061 1.19%
dense LU matrix factorization 79.82538155 75.48281949 5.75%
============== ============== ============== ============== ==============
Draw Canvas 58.95085 58.947132 fps 0.01%
Draw Circle 46.805817 36.890232 fps 26.88%
Draw Circle2 33.505802 32.472202 fps 3.18%
Draw Rect 16.730171 16.0976 fps 3.93%
Draw Arc 21.473064 21.261707 fps 0.99%
Draw Image 20.328669 19.308422 fps 5.28%
Draw Text 26.529566 25.859564 fps 2.59%
============== ============== ============== ============== ==============
OpenGL Cube 58.891674 58.88588 fps 0.01%
OpenGL Blending 62.611702 62.69004 fps -0.12%
OpenGL Fog 62.638985 62.57051 fps 0.11%
Flying Teapot 60.60055 60.60055 fps 0.00%
============== ============== ============== ============== ==============
Garbage Collection
Stretching memory binary tree of depth 16
Total memory 14290928 12582912 bytes -11.95%
Free memory 6946160 5074720 bytes 36.88%
Creating long-lived binary tree of depth 14 and long-lived array of 125000 doubles
Total memory 14290928 14290928 bytes 0.00%
Free memory 6255520 2128808 bytes 193.85%
Create 37448 trees of depth 2
Top down 360 453 msecs 25.83%
Bottom up 300 357 msecs 19.00%
Create 8456 trees of depth 4
Top down 296 372 msecs 25.68%
Bottom up 349 379 msecs 8.60%
Create 2064 trees of depth 6
Top down 352 350 msecs -0.57%
Bottom up 311 363 msecs 16.72%
Create 512 trees of depth 8
Top down 342 392 msecs 14.62%
Bottom up 339 372 msecs 9.73%
Total memory 14290928 14684144 bytes 2.75%
Free memory 5631192 5537968 bytes 1.68%
Completed in 2927 3333 msecs 13.87%
The 0xbenchmark is the one they used in the video whereas AnTuTu and Passmark are another couple I downloaded from the store. In 0x Linaro has made many clear improvements, however in the other set of benchmarks they are within the margin of error of each other but only barely (with Linaro in the lead). From this I'd say that it has promise but needs some more refinement before it can offer consistent improvements to Android. I feel that they relied too heavily on that one set of benchmarks (which has not been updated since October 2011 before Android 4.0.1 was even released) and they need to test it in other situations. Also for at least me I had to run Sunspider on my own, it would crash whenever I ran it through 0x, indicating to me the benchmark is in great need of refinement.
I don't mean to offend but it could also be partially the implementation of the Linaro in CM9. It was just released and I imagine a few iterations of real world testing are needed to perfect it.