Good point, this is on the way. The problem is more to do it right, that is, you want for each "word" to detect if this is a word, a chemical formula or a SMILES. Then, if SMILES, you then need to canonicalize it and search.
Nothing very complicated, but it needs to be well done to be of any use.
Plus, to detect if its a systemic, or semi-systemic name, and extract the structure from that. I know of three tools which do that, and only one is free.
If you want a SMILES detector, you can use my opensmiles-ragel grammar to detect if a word is syntactically correct. (Not grammatically correct; it will allow "c1C" unless you write code to require balanced parentheses and matching ring counts.)
This would be much faster than passing it to any of the cheminformatics toolkits to do the first level detection.
If you're doing similarity searches, you might be interested in my chemfp project.
Nothing very complicated, but it needs to be well done to be of any use.
Thanks for the feedback!