||Too much overhead. (create a new shapefile for each point?)
Here is the problem: I have AML scripts that start at the base of a river (FlowAccumulation output) and "walks" up the river channel by stepping into the cell with the next highest dicharge.
As the "cursor" walks up the stream it outputs to a text file the current elevation, river mile, slope and other attributes.
This type of query and response of the DEM allows a script to create new tools that ESRI hasn't thought about (whithout going through ArcObjects). Like new ways of generalizing data, smoothing, aggregating maybe you want to calculate wind fetch on a lake, or like in my example, creating a river profile.
None of this is possible without the ability to read individual pixel values efficiently. Ideally, we would also have the ability to post new data to an empty raster (or modify a dem) using a similar syntax.
Today I tend to dump grids to floating point binary files, load them into python as arrays and work with them that way. The results are put back at the end of the script. There is a lot of overhead associated with dumping and reforming the grids.
What I envision is some kind of efficient cursor into a raster dataset like you have provided for vector tables.