First of all, if you're reading from io.Reader consider it as reading from the stream. It's the single input source, which you can't 'read in parallel' because of it's nature - under the hood, you're getting byte, waiting for another one, getting one more and so on. Tokenizing it in words comes later, in buffer.
Second, I hope you're not trying to use goroutines as a 'silver bullet' in a 'let's add gouroutines and everything will just speed up' manner. If Go gives you such an easy way to use concurrency, it doesn't mean you should use it everywhere.
And finally, if you really need to split huge file into words in parallel and you think that splitting part will be the bottleneck (don't know your case, but I really doubt that) - then you have to invent your own algorithm and use 'os' package to Seek()/Read() parts of the file, each processed by it's own gouroutine and track somehow which parts were already processed.