Google Scholar's solution is to show a captcha when a given IP has made too many requests in a given time period, although a scraper can easily throttle to avoid this.

You could require a captcha every n page views, or you could render the text of the page as a distorted image, which would defeat the OCR approach others have suggested here. These would make scraping difficult, but they would also mean throwing UX out the window, and they could still be defeated with Mechanical Turk.