That is probably because the Evernote founders were the team that developed the CalliGrapher cursive recognition engine which was used by the Apple Newton back in the day.
Maybe, but it seems to me that for search (especially with stemming / misspelling coverage) you need a much lower OCR accuracy than for converting into text. Like maybe 95% vs 99.5% - you can have an order of magnitude more errors and search would still work fine.
