Methods do not get serialized so it would seem more straighforward,
clearer, more inutitive, etc. in reading the code to use structures.
I know in .Net, structures can have methods, but I've used them in the
past strictly for grouping relate elements/data types.

The differentiation I'm thinking of that may be pertinent to performance
is that structures are created on the stack versus on the heap for class
instances.  My guess is that for small sets of data, perhaps the
structure is better and the opposite is true for returning large sets of
data.  I really do not know.

I have not found any discussion of such in MSDN or otherwise.  I'm
curious to know if anyone has any remotely conclusive evidence or
experience.  Thanks.