I found that the new kernel has a weak performance.
The memory benchmarks (copy chars, memcpy ...) need twice the time over the 22 kernel.
Same for the NEON support. A benchmark provided by TI
http://wiki.davincidsp.com/index.php/Cortex_A8#Neon_and_VFP_both_support_floating_point.2C_which_should_I_use.3F
take 4 times longer.