Suggestion: use lua_createtable() when number of columns is known will improve performance.
Avoids multiple reallocations of the table as the columns are added to the result table. 

Due to current use of lua_newtable(),

for a1,a2,a3,a4,a5,a6,a7,a8 in stmt:urows() do 
   local row = {a1,a2,a3,a4,a5,a6,a7,a8}
end

is significantly faster than

for row in stmt:rows() do end

Egil, Good suggestion, thanks! This improvement [a56548c06d] will be in the next version.