So the Snapdragon 615 will not perform better than a Tegra 3? That's hard to believe.
I think stuff like...more efficient and accurate branch predictor, better blah blah blah, higher memory bandwidth, and larger L2 cache
will provide performance improvements that should show up in the majority of use cases, whereas the DMIPs measurement is like an absolute best case throughput measurement. (If I recall correctly when reading about ARM's response to that question...) and so yes, it should be even better clock/clock than Cortex A9.
but this still leaves my concern at the 'it doesn't clock as high' stage.
but in general I'm a naysayer on this topic and absolutely think that
little.little 4x1.7ghz/4x1.0ghz is just plain dumb. Either give me a 3ghz/1ghz option or go home. IMO.