Consider Type IIB string theory with 4 O7-planes and 32 D7-branes on $K3 \times \mathbb T^2/\mathbb Z_2$. The K3 induces D3-charge on their world-volumes which can be cancelled by the introduction of 24 D3-branes. Performing a T-duality transformation along the $\mathbb T^2$ one obtains one O9, 32 D9-branes and 24 D5-branes in Type I, provided the $\mathbb Z_2$ is chosen appropriately.
If I now go in the orbifold-limit of the K3 where it is represented by $\mathbb T^4/\mathbb Z_2$ I have the same geometry and amount of O9/D9-branes as the Gimon-Polchinski model. But in this model there are 16 O5-planes and 32 O5-planes, such that the six-form tadpole vanishes.
How is this consistent? Do the additional branes and planes appear when going to the orbifold-limit of the K3 or is it even possible to do this transition? Is there a mistake in my reasoning?