25 cm seems more than reasonable. The two linearly polarizing antennas are orthogonal, so by simple arguments of geometry, their far fields are orthogonal, and 25cm is sufficiently above a wavelength, so that we don't have to consider near field reactive coupling.
The exact formula why orthogonal fields do not couple is actually vector space math, so I'd go find a text book for higher math for engineers, first semester. That answers why if you believe me that from orthogonal geometry yield orthogonal field components, and from orthogonal field follows no interaction physically, you're all set.
If you want in on how the physics care about vector spaces and orthogonality, then get the text book from the same uni that goes into static fields and waves. Finally, you would want to arrive at Maxwell's equations and the bath needed for basics of finite element modeling to be able to simulate antennas, but you'd need to go through the math basics first. For an antenna as complex as this, you need to understand what you simulate, and simulate, to figure out what the near field is, as there's no closed form formula for complex geometric systems that you could reasonably calculate by hand.
It's also the number that the company that designed and simulated the antenna gave you - so I'd honestly just trust them on that. If the computer said "strength of the near field at the reflecting end at more than 12.5cm negligible", it is.
I knew instead it had to be two times the wavelength
That sounds like one of these rules of thumbs that are very rarely accurate. Never heard of that for such antenna constellations, so unless that $2\lambda$-rule comes with a very solid explanation and well-reasoned boundaries to what it applies to, I'd ignore that piece of "knowledge".