I agree with you that "true personalized search eludes them because they don't have enough data from you to tune up a model for you". That's what Larry Page said as well "Google doesn't know what you know". His ultimate goal is Answer Machine powered by AI but that's not happening anytime soon. I think internet search engines that we are using today are primitive compared to what we will have in the future.
Google's search algorithm is tuned up for searching the whole web. It turns out the heuristics you need are very different depending on the size of the collection.
When Gerard Salton was doing IR experiments with punched cards he was working with collections of as little as 70 documents and in that case you are going to be very concerned about recall and not precision. Maybe there is 1 relevant document and if you miss it you failed.
If you had 70 billion documents you might have 10,000 relevant documents and if you lost 60% of them you still have 4,000 documents. The end user gets more results than they can sift through.
Thus I always groan when I see a site is using "Google Site Search" because the relevance is usually worse than you'd get with the alternatives.
Connected with that is the tuning work: Google has sufficient data to tune up a big model for everybody but true personalized search eludes them because they don't have enough data from you to tune up a model for you.