dayyan is correct, it's reverse indexes which make search engines fast; there's no way to accelerate regex search over a petabyte of content if you only have 100 terabytes of flash disk. Keyword searches, reverse index, no problem.
blekko's web grep (https://blekko.com/ws/+/webgrep) supports regexes, but most of the searches we get for it are for constant strings, usually which are in the HTML, because that's what's interesting: who uses microformats? who uses various javascript libraries? who uses various comment systems? And so forth.
If you sent us a regex, we'd be happy to run it for you.
Running these searches consists of a MapReduce job run over all the html in our crawl. That's why it takes a while (a day or two) to get an answer.