What is the evidence for the existence of a Hubble sphere beyond which galaxies recede faster than light The title has been changed from the original title, because that is what I
meant, but I misused the terminology, and probably some physics, asking about
evidence for the existence of a horizon for the observable universe?
Though my question was a mess, Christoph put some work into giving it a cogent
answer, clarifying some points, pointing out errors, and adapting to what I really wanted, which I explained in a comment. As I was asking him, he
suggested however that I do not change much the question, which contains
probably common misconceptions.
With my thanks to him, here is the question as
originally asked:
From what I understand, this observable universe horizon results from the
expansion, so that when galaxies are sufficiently far away, the expansion
speed relative to us is faster than the speed of light. That is clean and
simple reasonning.
But, isn't it possible to imagine a geometric structure of the universe such
that the speed of light is never exceeded despite the expansion. Then there
would not be an observation horizon. 
After all, we found with relativty that speeds add in a more complex way that
the simple addition we were used to. Could it be that the simple expansion
analysis that we use to justify the existence of a horizon for the observable
universe could be refined and give a different result?
What we observe is not very far (since the universe is only 13,7 Glyr
old). More precisely, we observe it from a time when it was not so far (sorry
for the over-simplified statement), even though what we may observe is today
much further away. But there is no way to observe things that approach the
suggested distance of the horizon (around 80 to 90 Glyr, from various posts).
 A: 
From what I understand, this observable universe horizon results from the expansion, so that when galaxies are sufficiently far away, the expansion speed relative to us is faster than the speed of light. That is clean and simple reasonning.

Maybe that's clean and simple, but it's also wrong. If you mean recession velocities by "expansion speed relative to us", those reach $c$ at the Hubble sphere instead of the event horizon, and we do see objects more distant than that. A better fit would be relative velocities as computed by parallel transport along the light trajectory - except that sources outside the event horizon are causally disconnected from us, ie there is no null geodesic connecting us with them and hence no way to compute relative velocities.

What is the evidence for the existence of a horizon for the observable universe

It's a prediction of our model, sensitive to the time evolution of the scale factor $a$. We think the end of the universe will happen at a finite conformal time
$$
\eta_\text{max} = \int_0^\infty\frac{dt}{a(t)} \lt\infty
$$
The cosmic event horizon is the past light cone at that time. In terms of conformal time $\eta$ and (appropriately defined) comoving radial distance $\chi$ it's given by
$$
\chi(\eta) = \eta_\text{max} - \eta
$$
Light emitted by sources beyond that boundary would take a 'longer than infinite' cosmological time to reach us.
If the time evolution of the scale factor were such that the integral above wasn't finite, there wouldn't be a horizon, and the observable universe would continue to grow unconstrained instead of asymptotically.

Regarding your clarification, the Hubble sphere doesn't have that that much physical significance. It's the place where recession velocities reach $c$, but those aren't true relative velocities, but apparent velocities based on a somewhat arbitrary space-time-decomposition at a distance.
Take the galaxy GN-z11, the most distant astronomical object with spectroscopically determined redshift, with $z\approx11$. When the light we receive today was emitted, the galaxy had a recession velocity on the order of $4c$, and is still receding 'today' (ie at current cosmological time) with a velocity of more than $2c$. Nevertheless, we're able to see it - though we'll never be able to see it as it exists 'today' as it has already crossed the event horizon a couple of billion years ago.
How can that be, you may wonder? For an analogy that isn't really exact, but might get across some of the features involved, consider the following situation in special relativity:
You're sitting on a space station. Two ships launch in oppostite directions, with velocities of $0.6c$ each. From your perspective, the distance between the ships grows at a rate of $1.2c$. And yet, the ships are able to send light signals between each other due to the invariance of the speed of light: From your perspective, the signal starts out with a velocity of $c$ despite the emitting ship's motion, and hence can catch up to the other ship; there also isnt really anything special about ships moving with $0.5c$ each so their apparent recession velocity adds to $c$.
A better analogy could propably be created using accelerating ships (as this will involve Rindler horizons), but I'd have to work out the details first.
