Send a request to Google's geocoding service and you can get back a single address. There is a limit of 50,000 in a day. Since it has to go off across the internet it may not be faster than using in_memory featureclasses (that is not a real file). Did you miss the clever trick suggested by Chris Fox using in_memory as a virtual disk drive?I assume that if you want a fast response from one address you are trying to geocode a lot of them? If so, take out the loop and geocode them all in one featureclass that will be really fast. You should expect 1 million per hour if done locally on real files.Here is a way to use Google:
# geopy example
# http://code.google.com/p/geopy/wiki/GettingStarted
from geopy import geocoders
apikey = "your_api_key_adfvaklfmuyrdghbndfzcvazdfzcvsfgsfvsfgsdfbvsfgsghsdfgd"
g = geocoders.Google(apikey)
place, (lat, lng) = g.geocode("380 New York Street,Redlands,California,USA")
print "%s: %.5f, %.5f" % (place, lat, lng)