First Some Definitions
Before I answer, let's first make a subtle nomenclature issue clear to check that I'm understanding your definitions: a set $\mathcal{T}=\{X_1,\,X_2,\,\cdots,\,X_M\}$ of Lie algebra vectors is called a set of topological generators for a group $\mathfrak{G}$ iff the closure of the smallest group containing elements of the form $\exp(\tau\,X_j)$ for $\tau\in\mathbb{R}$ and $X_j\in\mathcal{T}$ is the group $\mathfrak{G}$, i.e.
$$\mathfrak{G} = \overline{\left<\{e^{\tau\,X}|\,\tau\in\mathbb{R};,\,X\in\mathcal{T}\}\right>}$$
where, of course, there is implicitly some supergroup either equal to or containing $\mathfrak{G}$ whose group topology we use to define our notion of closure.
By this definition, the Lie algebra generator $(i,\,i\alpha)$ of the one-parameter irrational slope subgroup $\{(e^{i\,\phi},\,e^{i,\,\alpha\,\phi})|\,\phi\in\mathbb{R}\}$ of the torus $\mathbb{T}^2=\{(e^{i\,\phi},\,e^{i,\,\theta})|\,\phi,\,\theta\in\mathbb{R}\}$ when $\alpha$ is irrational is itself a topological generator of the whole torus.
Another definition which I think will be useful in this context is where $\mathfrak{G}$ is the smallest group containing the exponential of the smallest Lie algebra $\mathfrak{g}$ containing $\mathcal{T}$. This is a stronger concept that topological generation, and I haven't seen a name for it. Let us call it simply "connected immersed Lie subgroup generation" (CILSG) or "connected analytic subgroup generation" (there are some authors who do not call a Lie group that is immersed without being topologically embedded in a supergroup a "Lie subgroup" but rather an "analytic subgroup"). In this case $\mathfrak{G}$ is the group of all finite products of elements of the form $\exp(\tau\,X_j)$ for $\tau\in\mathbb{R}$ and $X_j\in\mathcal{T}$ is the group $\mathfrak{G}$:
$$\mathfrak{G} = \left\{\left.\prod\limits_{j=1}^M\,e^{\tau_j\,X_j}\right|\;X_j\in\mathcal{T};\;\tau_j\in\mathbb{R};\;M\in\mathbb{N}\right\}=\bigcup\limits_{k=1}^\infty\,\exp(\mathfrak{g}_s(\mathcal{T}))$$
where $\mathfrak{g}_s(\cdot)$ means "smallest Lie algebra containing the argument set of vectors".
By this definition, the CILSG of the the Lie algebra generator $(i,\,i\alpha)$ is simply the one parameter subgroup $\{(e^{i\,\phi},\,e^{i,\,\alpha\,\phi})|\,\phi\in\mathbb{R}\}$. We don't form its closure.
I shall only address connected immersed Lie subgroup generation (CILSG) in the following will not address strict topological generation further because the question of whether a particular subgroup is closed or not gets a bit messy and is highly particular to a given example. Basically what you need to do is test whether all the possible one parameter subgroups of a given group are topological embeddings. Then the Lie group in question is an embedding (this is much harder to apply than its sounds). See section 2.3 "Virtual Lie Subgroups" as well as Theorem 5.4 in:
V. V. Gorbatsevich, A. L. Onishchik and E. B. Vinberg, "Foundations of Lie theory and Lie transformation groups", Springer-Verlag, 1997
Instead, my CILSG definition is the image of the smallest Lie algebra $\mathfrak{h}$ containing the generators in question under what Rossmann calls the functor $\Gamma$ inverting $\mathrm{Lie}$ in the Lie Correspondence (the one to one correspondence between Lie subalgebras $\mathfrak{h}\subset\mathfrak{g}$ of the Lie algebra $\mathfrak{g}$ of a Lie group $\mathfrak{G}$ and its analytic (immersed Lie) subgroups. See Section 2.5 of
Wulf Rossmann, "Lie Groups: An Introduction Through Linear Groups", Oxford Graduate Texts in Mathematics
The Answer
With these definitions, we have the following:
The smallest Lie algebra containing three linearly independent "infinitessimal" boosts is the whole of the Lie algebra $\mathfrak{so}(1,\,3)$. Accordingly, the group comprising finite products of members of the one parameter subgroups generated by three boosts in linearly independent directions is the whole of the identity connected component $SO(1,\,3)$ of the Lorentz group. In other words, it is the proper, orthochronous Lorentz group. In this case, the generators are also topological generators, because the closure of $SO(1,\,3)$ in $O(1,\,3)$ is $SO(1,\,3)$ itself.
The smallest Lie algebra containing two linearly independent "infinitessimal" boosts is the Lie algebra spanned by these two boosts together with an infinitessimal rotation within the plane spanned by the two directions (i.e. about an axis normal to this plane). Accordingly, the group comprising finite products of members of the one parameter subgroups generated by two boosts in linearly independent directions is an isomorphic copy of the connected component of the group $SO(1,2)$.
I cannot answer whether the closure of the group in 2. can ever be the whole of $SO(1,\,3)$; I suspect that it cannot be.
Where Are These Things Written Down?
Whereas it seems to be fairly widely known that the set of all finite products of exponentials of a given set of Lie algebra members is precisely the connected Lie group with the smallest Lie algebra containing the set under the Lie correspondence, this is a surprisingly hard piece of information to find written down: I too have had your experience.
I have written up a proof in section 2 of my paper below on ArXiv, but it has not yet been peer reviewed. I hope it shall be soon in the form of a book that I am publishing. But until that time, this is the best I can offer. It is not a hard or subtle proof, so I have high confidence in it and hopefully you'll be able to see what's going on too:
R. W. C. Vance, "Fine Entanglement and State Manipulation of Two Spin Coupled Qubits: A Lie Theoretic Overview" arXiv:1502.05200 [quant-ph]
