What should be done is start with the intuition and visualization, and then show how the ideas can be made precise. If you would have been taught only the epsilon-delta form from the start, it would have been easy precisely because you would not yet have obtained the real understanding.
Even in high dimensional spaces visualization is very important. This often happens by analogy with lower dimensional spaces. For example if you start with a region in Hamiltonian phase space, then the region's volume is constant as time evolves. This is a highly intuitive and visual statement. Saying the same in symbols wouldn't be nearly as clear.
Additionally, some things like ordinary algebraic manipulation are very well-suited to linguistic abstractions ("multiply the polynomials, take the derivative, put all terms involving z on one side of the equation, apply the quadratic formula"). Sometimes only the linguistic abstraction can give the solution (e.g., "this problem is easy because the quadratic coefficient cancels out, and the equation is in the form t^3 + c t = d").
It's also worth noting that manipulating the linguistic abstractions takes a lot of insight and talent (e.g., knowing the perfect substitution of variable to make an integral fall into a known form, or knowing which one of the four error terms will be hard to control, and working on it first).
It's not wise to be over-committed to the visual approach.
The advantages of thinking this way seem to be a little like the advantages of test-driven development: time spent understanding representative concrete cases doesn't teach you everything, but it can eliminate many misunderstandings very quickly.
That said, I believe the above is for the people who will never get to higher level mathematics, but rather are struggling with first level calculus.
Visualization will not get you beyond 3 dimensions, nor will it get you understanding systems in terms of Lagrangians/Hamiltonians, nor will it give you the ability to read texts geared toward actual mathematicians.
Speaking for myself, it was surprisingly difficult un-learning the "slope of a tangent line" type of conceptualizing in order to understand math with sufficient rigor to be able to actually read math texts correctly.