Updated on 8 May 2008 to account for new data and changed methodology.
I have updated the attached placemark collection at the top of this post with a new version that corrects some coding errors and enhances the usefulness of the data by highlighting the top polluters.

Key changes:

• A red marker is used for the top polluters and the top emissions to air, water and land are highlighted in bold.
  
• Where an emission of a substance to air, water or land is greater than the 95th percentile (assuming a lognormal distribution) I have noted the site as a top emitter. There are 549 air sites, 81 land sites and 51 water sites (with some overlap, a total of 600).
  

It is worth noting that the coordinates provided by the NPI are not always accurate. To improve this I used the Google Maps API to geocode all the premises that had an address. Where there was a street name and number, or an intersection of two streets I substituted the geocoded coordinates. 1256 of 3951 sites were geocoded in this way.