Grabbit should be the quickest. Is there any specific issue that you are facing setting it up ?

For geocoding, you would need to use google maps like service, but for performing geospatial search you can use Solr.

        You can write a event listener service, which is run mode aware. You can can listen to relevant jcr events and trigger content flush from publish. You'll have to setup flush agent on publish .instances.

        Not sure I understand the issue you are facing with dispatcher cron job. Do you mean that when the cron runs on dispatcher, publish might not have generated a new version of the page and dispatcher would pull the older version again?

        You can write a cron job which clears cache once a day.

Dispatcher does have directive to ignore url parameter and cache the page in question even if the incoming URL has query parameters. Now personalization in this scenario will only work if the personalization is implemented client side (teasers should work as they are executed client side.).But if yo...

        You can use query string, with the understanding that it impact cacheability of the content in question and impact of this decision on performance. Not sure if  it works for your problem, but a webserver rule can be defined which converts query string into a url hash parameter. This was you ...

       Instead of using query string, you can also use url hash strings and map url hash key value pairs to segments.

You can update the host file on your machine to point localhost to the domain you want tho use.

I am assuming you are using something like this to calculate the distance - http://en.wikipedia.org/wiki/Haversine_formulaI don't think JCR has support  for trignometric functions to do radius search, if thats what u want to do.