Nice article and thank you for putting together the comparison.

One of the reasons that any embedded solution will be faster is related to the overhead of client/server communications... There is a penalty for every connection to SQL Server. For small record sets and optimized code it's very hard to overcome that penalty. I'm curious if instead of 5000 rows if your query were a million rows (Not that much today) would SQL come back out on top?

Of course another reason that embedded solutions are faster has to do with SQLserver being optimized for Muti-User, To compare to an embedded solution the database should be put into Single-User mode.