Furthermore

On the performance front, we quickly realized that the JavaScript-based markers often used for such mashups just would not work for a large data set such as this one (80,000 registrations and counting). So for our high-level views, we use the Java Image I/O API to create our own matrix of tile images, and then use Google Maps API calls to have those overlaid atop the normal map tiles. For example, where a fully zoomed-out view of the earth would cripple one’s browser with the work of placing 80,000 individual “balloons” on a map using JavaScript-based markers, the custom tile approach means the browser just has to retrieve and position at most a handful images. Much faster and kinder to your browser.

My AJAX-slinging Ruby-on-Rails riding, java-is-so-last-century developer wife says :

it’s called “grouping” you knob. Maybe you should be checking out mapstraction.