I have two import processes in place, one something of a bootstrap, using VSTO, that provides a lot of interactivity with the spreadsheet, allowing for good feedback to the user. The other is ODBC based, and it simply treats a spreadsheet tab as a table, spreadsheet columns as table columns, interacted with using SQL -- but the user has to provide the worksheet name, because ODBC can't provide a collection of worksheet names (as far as I'm aware). This is one feature (the ability to choose from a list of worksheets) I provide through VSTO that I'd like to see in the web-based import, and not just the bootstrap desktop client. On top of that, it would be great if I could alter the original spreadsheet with error-coded colors, and spit the whole sheet back to the user -- DEFINITELY something ODBC won't allow. Right now we still feel that the number of/potential for errors in the spreadsheet is still beyond the realm of documentation and training and templates alone to handle, and limited feedback doesn't help.
Is the only downside to actually using VSTO on the server the Excel app overhead, or is it a licensing issue? also, can the alternative approach you mention be used over the Web? we're not talking about local network operations here.