First of all, google has had a build-in calculator/unit converter for a very long time.
Second of all, I think that an endeavor like this is definitely doomed until computers can at least precisely understand long paragraphs of text (which are what is needed to specify the meaning of more complex searches). This comes from the simple fact that a simple one-sentence "natural language" question can often be interpreted in a bunch of different ways, depending on context and many other things. The only thing that a search engine like this can do then is to choose the
most common meaning for the sentence, which will mean that while this search engine can be good for Joe Blow, it will suck and only return irrelevant results for most more advanced searches. It can of course also choose to display results for
all possible interpretations of all given questions, but this will result in hundreds of pages of results to sift through before finding anything remotely relevant what you really wanted to find.
After all, what a search engine like this does in the end is to translate a "natural language" sentence to the correct set of exact search terms, which is a kind of "compression". As has been proven in computational science, only a "simple" enough subset of data (which often means "common"

in a certain data space can be successfully compressed with a given algorithm, while there must always be another (larger) part of the data set that cannot be compressed (i.e. the more "uncommon" data). This uncompressable data will even become
bigger than its original form when you attempt to compress it, which in this case manifests itself as half a natural language essay to describe a more complex search. Not only is it harder to formulate and type in this essay than just the corresponding pure search terms, but as mentioned at the beginning of this post, there will probably be a long time before a search engine can make anything useful out of a "question" in the form of a larger text mass like this.