Very good point. I think spreadsheets are just deep enough and map well enough to the domains that use them, that it can tread water. (Not swim, mind you.) They're also quite clear about their constraints.
Of course, from a software engineering perspective many things that are developed using spreadsheets are truly horrific - but the non-developers who create complex systems using them love them.
Spreadsheets have a much more brilliant potential than we currently understand. In fact, it's my opinion that in the future we'll be essentially coding in a spreadsheet; not a text editor. (That's the case with the project I'm working on.)
If you want to talk about this more, I'm david927 at gmail.
Simon Peyton Jones et al. wrote an interesting 2003 paper on extending Excel's "natural" mapping to functional programming with first class (i.e. cell-based) user-defined functions.
If less of our finance system depended on spreadsheets and were actually well understood and observable by properly designed software, the 2008 crisis could have been averted.
Doesn't seem to have stopped Excel in particular, and spreadsheets in general, from being wildly succesful.