Wolfram Alpha is trying to do the same -- trying to give really good answers/computations for common sorts of queries.
Users expect reasonable answers for all queries.
If they can get their content to a respectable level for any given subject area, eg physics, I can see Wolfram Alpha becoming a valuable tool, worth paying a reasonable subscription to use.
No, it doesn't compete with search engines, but it will fail for the same reason that Microsoft search failed as explained above.
But your suggestion of WA as an ad makes sense - but what is held back in this freemium hypothesis, that will make users need Mathematica?
I suspect that Microsoft's failure in this regard would play out in many other contexts including this one. Imagine for example if Mathematica were to have a much higher failure rate.
Mathematica works as well as it does because an enormous amount of research has been put into one domain. Wolfram Alpha is not like that at all.
It's a matter of expectation.
And it seems likely that creating this expectation is a deliberate ploy on WA's part to create publicity. "Google killer" is a much better media story than "mathematica online". They've designed the front page in a google-style, with a similar text interface, and probably many other ways I haven't noticed.
That being said, I don't think it's particularly fair to judge this genre of product on the first day of its initial release. I believe that over time, the developers will likely improve the algorithm.