The thing is, "better cpu" is only part of the story: a more powerful SoC (System on a Chip) can be negated by poorly-optimised software, and a benchmark app will not tell you that because it isn't testing UI performance or RAM management when multitasking. Then it also depends on what you use the phone for, e.g. someone who spends 30-60 minutes at a time gaming will care about how the phone manages thermal load over that time, which a 1 minute Antutu run won't tell you (you can get a high graphics score in the 20 seconds or so it spends testing that, but if the phone throttles the gpu horribly after 10 minutes of gaming that benchmark score will have been very misleading). Other people who mainly use their phone for messaging and watching the odd video won't care about that at all. This is why I say that it's the performance of the device in your real usage that matters, not just some measure of what the SoC can do in artificial conditions.
And that's leaving aside the fact that many manufacturers were caught "gaming" the benchmarks, i.e. programming the devices to behave differently (e.g. run all clocks flat out regardless of temperature) when they detected that popular benchmarking apps were being run - making those benchmarks even less representative of real performance.
So it's a complex problem. Even accepting the limitations of benchmarking apps, the "overall" score isn't that useful because that mixes many things together; so the phone may get a high score due to a test of something that's not really a limitation in real use, leading to a decent "overall" score, but that may hide a poor performance on something that's important to you. Doing separate tests for things like speed of reading/writing to storage (sequential or random, which are not always correlated the way you would guess), single/multi-core computation tests, web browsing, frame rates in graphics benchmarks as a function of run time gives better information - look at something like an Anandtech review to see many different tests done on different aspects of phone performance. But at the end of the day even these are only imperfect indicators of how the device will perform in real use. People use them in reviews because they give quantifiable scores, which are easier to standardise and compare than real usage, but what's easily measurable and what actually matters are not necessarily the same thing. This is why I insist that the real test is "does it do what I want adequately" rather than anything else.
) replies about Antutu EVER 
