I notice no marketing mentions MIPs, and its not mentioned in specifications for processors.
How reliable is this statistic? By now we all know more Mhz and Ghz is not necessarily a good direct comparison
They're not good. Modern architecture makes MIPS an antiquated measurement. If you're looking to programmatically determine the speed of the running system, it might give you a ballpark on the generation of the machine, but the best measure if you're shopping for a machine are the big benchmarking sites. e.g. http://www.cpubenchmark.net/high_end_cpus.html
There's a article on Wikipedia that discusses some of the weaknesses: http://en.wikipedia.org/wiki/Instructions_per_second
It depends what you want to measure but in general it's not a practical real world measurement for general bechmarking and it's hard to compare across different architectures - for example, my processor might be fantastic at executing a NOP (no-operation) instruction in a loop, but it might be crap at shift operations or my SIMD microcode may be sub-par and thus no good for heavy graphics, digital signal processing or encryption/decryption.
MIPS is a bit like comparing the performance of a group of cars and vans by putting them all in neutral and revving the engines to max and concluding that the Porche must be best - only to find that the group test is to pick a vehicle for carrying freight.