Well, that's deliberate.
If I have a series of, say, exp(x) at 0 and another one of cos(x) at 0 and you ask GiNaC to compute them to order N, then you get two series ending with the same Order(x^N), which you can add, zip, and compare term-by-term. In contrast, asking for "the first N orders" would get you about twice as far for cos(x) as for exp(x).
I understand this use case, but ours is different. We have integrals that need to be expanded to a given order; each integral has a prefactor; we have our method of expanding integrals without prefactors, but we need to know the orders in advance, so knowing the leading order of the prefactor is an essential first step.
This was done for computations in dimensional regularization where one wants to make sure to catch all terms up to some pre-known order.
Maybe you could explain your use-case a bit more? Are you really interested in "the first N orders" or just "the leading order"?
Just the leading order is sufficient.