I don't understand the problem, either. There is no "strip" operation on Windows that could cause this problem; relocation information is always present and cannot be removed under normal circumstances. So I think the problem is elsewhere.

If Josh Ritter could report what precisely he did to determine that the 2.5-shipped version was "stripped" and the sqlite.org version isn't, that would be helpful.

According to http://support.microsoft.com/kb/q196069/
the error 998 of LoadLibrary indicates ERROR_NOACCESS if an access violation occurred, 
so that might be a possible cause. To analyze this further, one would have to run
the application under a debugger.