Like any feature, the usage is almost entirely dependent upon the quality of the implementation. It is very difficult to create a good recommendation engine; if you manage to do so, and repeat visitors find it useful, they will use it.
Anecdotally, I have used these features in Amazon and NewEgg. I have found little value in YouTube recommendations, because my noise ratio is high (I look at lots of videos that were linked from sites like Reddit, but few on my own). But I do look at the feature, because it is prominently placed despite the low quality of the results.
I would strongly suggest creating your site so that a recommendation is entirely optional to the experience. When you have less data to work with, the recommendations will be of low quality. Once you have a breadth of data to mine on your users, you should (note that the 'should' involves a whole lot of 'it's hard, but doable') be able to create a feature like this useful to your visitors.